An AI personality generation method and system for child adaptive companion interaction

Abstract

The application discloses an AI personality generation method and system for child adaptive companion interaction, relates to the technical field of artificial intelligence interaction, and comprises the following steps: collecting multi-dimensional original data of a child, performing time window aggregation and feature calculation, obtaining a child instant state vector and a current interaction event type label; according to the child instant state vector and the interaction event type label, performing synchronous incremental updating based on the current interaction event, and outputting an updated child growth portrait and a robot life portrait; performing bidirectional coupling mapping operation according to the updated child growth portrait and the robot life portrait; performing a decision mechanism to obtain an interaction intention, an interaction strategy and an emotional expression intensity, and generating final interaction content and instruction operation; the technical problem that the learning achievements of children cannot be structured and stored to drive the change of robots is solved, and the technical effect of double-track traceable recording of the cognitive growth of children and the state change of robots is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence interaction technology, and in particular to an AI personality generation method and system for adaptive companionship interaction for children. Background Technology

[0002] With the rapid development of artificial intelligence technology, children's educational and companion robots are gradually entering the home environment, becoming an important tool to assist children's learning and growth.

[0003] Currently, children's educational companion products on the market can be mainly divided into the following categories: The first category is traditional learning tools, with content playback, practice questions, photo-based question searching, and voice Q&A as their core functions. Essentially, these are tool-type devices. Aimed at completing predetermined tasks, the products themselves lack stable personality traits and a sustainable growth trajectory, making it difficult to establish a long-term emotional connection and companionship with children. The second category is intelligent dialogue products, which can engage in dialogue and Q&A with children via voice. However, their companionship logic is usually based on a command-response Q&A model or content recommendation mechanism, lacking a clear shared growth relationship. The device's capabilities do not substantially change as the child grows. The third category is virtual pet development products, whose growth mechanisms mostly rely on entertaining feeding, clicking, or gaming behaviors. The technological coupling between the child's actual learning achievements and the virtual pet's growth is weak.

[0004] In summary, existing products generally suffer from unstable personalities. Companion robots, with prolonged use, easily become generic question-and-answer interfaces, lacking a coherent trajectory of personality development. Software upgrades typically only involve adding functionality; children don't perceive that the robot has truly grown up or understands them better. Current technology lacks an adaptive companion interaction solution that can deeply couple a child's genuine learning and growth with the robot's personality and embodied abilities evolution. Summary of the Invention

[0005] This application provides an AI personality generation method and system for adaptive companionship interaction with children, which solves the problem in the prior art that children's learning behavior and robot ability growth cannot be deeply coupled and interacted, and achieves the technical effect of driving the synchronous evolution of robot personality and embodied ability through children's real learning behavior.

[0006] This application provides an AI personality generation method for adaptive companionship interaction with children, including: S1: Collect multidimensional raw data of children to form an interaction record vector. Based on the interaction record vector, perform time window aggregation and feature calculation to obtain the child's real-time state vector and the current interaction event type label. S2: Based on the child's real-time state vector and the interaction event type label, and taking this interaction event as the basis, perform synchronous incremental updates and output the updated child growth profile and robot life form profile; S3: Perform bidirectional coupling mapping calculations based on the updated child growth profile and robot life form profile, and output a skill node unlocking instruction set; update the robot life form state model using the unlocking instruction set; the robot life form state model includes a causal response layer and a derived statistics layer. S4: Based on the child's immediate state vector, the updated robot life form profile, the derived statistical layer fields and personality parameters in the updated robot life form state model, execute the decision-making mechanism to obtain the interaction intention, interaction strategy and emotional expression intensity. S5: Receive the skill node unlock instruction set, interaction intent, strategy type and emotional expression intensity, and execute the three sub-tasks of skill unlock execution, achievement judgment and interaction content synthesis in sequence to generate the final interaction content and instruction operation; The multidimensional raw data includes a sequence of frontal facial images, three-dimensional spatial coordinates, voice signals, touch event data, relative motion data, and timestamps; the child's real-time state vector is a sliding window formed by taking all interaction record vectors within the previous 30 seconds as the endpoint, and calculating the child's state index; the state index includes focus score, fatigue score, emotional polarity score, interaction initiative score, and physical activity intensity score.

[0007] An AI personality generation system for adaptive companionship interaction with children, applied to the above method, the system comprising: The perception and recognition module is used to synchronously collect multi-dimensional raw signals from children within a preset sampling period to form an interaction record vector. Based on the interaction record vector, time window aggregation and feature calculation are performed to obtain the child's instantaneous state vector and the current interaction event type label. The multi-dimensional raw data includes a sequence of frontal facial images, three-dimensional spatial coordinates, voice signals, touch event data, relative motion data, and timestamps. The child's instantaneous state vector is a sliding window formed by taking all interaction record vectors within the previous 30 seconds as the endpoint, and calculating the child's state indicators. The state indicators include focus score, fatigue score, emotional polarity score, interaction initiative score, and physical activity intensity score. The profile update module receives the child's real-time state vector and interaction event type label from the perception and recognition module. Based on the child's real-time state vector and interaction event type label, and taking the current interaction event as the basis, it performs synchronous incremental updates and outputs the updated child growth profile and robot life form profile. The coupling mapping module is used to receive the updated child growth profile and robot life form profile output by the profile update module; read all preset skill nodes in the robot skill tree; perform bidirectional coupling mapping operation based on the updated child growth profile and robot life form profile; output skill node unlocking instruction set; update the robot life form state model using the unlocking instruction set; the robot life form state model includes a causal response layer and a derived statistics layer. The personality arbitration module receives the child's real-time state vector output by the perception and recognition module, the updated robot life form portrait output by the portrait update module, the derived statistical layer fields in the updated robot life form state model output by the coupling mapping module, and a set of configurable personality parameters; it executes the decision-making mechanism to obtain the interaction intention, interaction strategy, and intensity of emotion expression. The interaction generation module receives the skill node unlocking instruction set output by the coupling mapping module and the interaction intent, strategy type, and emotional expression intensity output by the personality arbitration module; it then executes three sub-tasks in sequence: skill unlocking execution, achievement judgment, and interaction content synthesis to generate the final interaction content and instruction operation.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: By incrementally updating the target dimensions in the child's growth profile through the profile update step, and simultaneously updating the robot's life form profile, the child's learning progress and the robot's state changes are recorded bidirectionally within the same interaction cycle. This solves the technical problem in existing technologies that children's learning outcomes cannot be structured and stored to drive changes in the robot, and achieves the technical effect of dual-track traceable recording of children's cognitive growth and robot state changes.

[0009] By using a coupling mapping step, the current mastery level of each dimension of the child's growth profile is compared with the mastery threshold of preset skill nodes. At the same time, the robot's current level is compared with the level threshold of the skill node. Only when both conditions are met simultaneously, the corresponding skill node is unlocked and the causal response layer of the robot's life form state model is updated. Meanwhile, the derived statistics layer is refreshed periodically. This ensures that the unlocking of new robot skills strictly depends on the simultaneous fulfillment of two independent conditions: the child's actual ability in the corresponding learning area and the robot's long-term accumulated level. This achieves the technical effect of making the child's real learning behavior the sole driving path for the robot's embodied ability improvement, fundamentally eliminating the possibility of acquiring ability growth through non-learning behaviors.

[0010] Through the comprehensive demand index calculation and three-layer decision-making mechanism of the personality arbitration process, the interaction intention decision, interaction strategy selection, and emotional expression intensity are all adaptively generated based on the child's immediate state, dual-profile data, and derived statistical layer data. This achieves the technical effect of the robot interacting with the child at the right time, with the right strategy, and with the right emotional expression. This constitutes a complete closed loop from perception to decision-making to expression, enabling the robot to adaptively generate personalized interactive responses based on the child's real learning events. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of an AI personality generation method for adaptive companionship interaction for children, as described in an embodiment of the present invention. Figure 2 This is an architecture diagram of an AI personality generation system for adaptive companionship interaction for children, as described in an embodiment of the present invention. Detailed Implementation

[0012] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0014] Example 1: As Figure 1 As shown, an AI personality generation method for adaptive companionship interaction with children is disclosed, the method comprising: S1: Collect multidimensional raw data of children to form an interaction record vector. Based on the interaction record vector, perform time window aggregation and feature calculation to obtain the child's real-time state vector and the current interaction event type label.

[0015] The robot uses a high-definition vision camera in its head, a DTOF area array chip in its chest, a microphone array, touch sensors, an IMU inertial measurement unit, and a rear-view ultrasonic sensor to synchronously collect multidimensional raw signals from the child within a preset sampling period. The multidimensional raw signals are then processed through analog-to-digital conversion, calibration, filtering, and feature extraction to obtain multidimensional raw data. Data is collected at a fixed sampling rate before and after each interactive session to obtain raw data; the new data replaces the corresponding old data, resulting in real-time data updates.

[0016] The multidimensional raw data includes, but is not limited to, frontal facial image sequences, three-dimensional spatial coordinates, voice signals, touch event data, relative motion data, and timestamps.

[0017] Specifically, a sequence of frontal facial images of children is acquired using a visual camera. This includes: extracting the coordinates of key facial points; calculating the child's current head yaw and pitch angles, both in degrees, with values ​​ranging from (-90, 90) to (-60, 60). Simultaneously, facial motion unit intensity values ​​are output to characterize the intensity of expressions such as smiling, frowning, and opening the mouth. The motion unit intensity values ​​range from 0% to 100%, and are dimensionless. The intensity value describes a relative proportion of change, not an absolute physical quantity. It measures the relative degree of muscle movement relative to a resting state (or relative to the individual's maximum motor capacity), rather than using physical units. While the absolute amplitude of facial muscle movement may differ between individuals, the intensity values ​​are normalized to eliminate individual differences, making cross-individual expression intensities comparable.

[0018] The three-dimensional spatial coordinates of the center point of the child's chest relative to the robot's body coordinate system are obtained by using a DTOF sensor, and the straight-line distance between the child and the robot is calculated in millimeters.

[0019] Children's speech signals are collected using a microphone array, and speech recognition and detection are performed. Valid text is detected, and the average energy value and speech rate estimate of the speech segment are extracted. The energy value is a dimensionless value, normalized to the range of [0,100]. The speech rate is measured in syllables per second.

[0020] The touch sensor records the type and duration of children's touch events on the robot's head, back, and other areas, and outputs the touch intensity level and the duration of a single touch. The touch intensity level is divided into 1 to 5 levels, and the duration of a single touch is in seconds.

[0021] By detecting the robot's own posture angles and motion state using an IMU, and combining this with the rate of change of the child's relative position (calculated based on three-dimensional spatial coordinates), the robot estimates the child's movement speed and activity intensity in the interactive space, with the speed measured in centimeters per second.

[0022] The current absolute time is read from the internal clock as a timestamp to determine the current time period, such as early morning, morning, afternoon, evening, or night.

[0023] After being timestamped, the above six types of raw data are merged into a single structured interaction record vector. This vector contains the instantaneous values ​​of all raw parameters and their corresponding acquisition timestamps. In this embodiment, the naturally occurring, unstructured, multimodal interactions between children and robots are transformed into numerical data elements that can be quantitatively processed, serving as a reliable source of raw data.

[0024] In this embodiment, the child's real-time state vector is a sliding window composed of all interaction record vectors within the previous 30 seconds, with the current time as the endpoint; within the sliding window, the child's state indicators are calculated. The state indicators include focus score, fatigue score, emotional polarity score, interaction initiative score, and physical activity intensity score.

[0025] In some embodiments, the attention score is taken as the absolute values ​​of the child's head yaw angle and pitch angle within the window, the average of the two values ​​is calculated on the window, and then the values ​​are added together and normalized to the range of 0 to 100. A higher score indicates that the child's face is facing the robot for a higher percentage of the time and that the child's attention is more focused. Its function is to quantify the child's current level of engagement in the interactive activity and to distinguish between effective learning periods and mechanical response periods.

[0026] The fatigue score is calculated as the average blink rate of children within the window, measured in blinks per minute. This is combined with the average silence duration between two consecutive valid verbal responses, and the weighted average is mapped to a range of 0 to 100. A higher score indicates a lower current energy level for the child, and a state closer to needing rest. Its purpose is to identify whether the child has exceeded the effective learning endurance boundary, providing a basis for the personality arbitrator to determine whether to recommend rest.

[0027] The emotion polarity score is calculated as the intensity ratio of smiling-related units to frowning-related units within the facial action units in the window, normalized and mapped to the interval [-50, 50], where a positive score indicates that positive emotions dominate and a negative score indicates that negative emotions dominate. This transforms children's ambiguous facial expression signals into quantifiable emotion tendency indicators for assessing emotion expression decisions.

[0028] The interaction initiative score is calculated by dividing the number of times a child initiates voice commands and touches the robot within the window by the total number of interaction events within the window, and mapping it to a range of 0 to 100. A higher score indicates that the child is more of an initiator than a passive responder in the current time period. Its function is to distinguish whether the child's participation is active exploration or passive coping, which directly affects the weighting of growth increments in the coupling mapping.

[0029] The physical activity intensity score is calculated by taking the mean and variance of the child's spatial movement speed within the window and mapping it to the range of 0 to 100. A higher score indicates more frequent physical activity, and its function is to help determine whether the child is in a state of excitement or anxiety. The above five indicators together constitute a five-dimensional child instantaneous state vector, with each dimension being a dimensionless score.

[0030] The weighted calculation (initial weight average division) and normalization methods used in this application are all common techniques in the field and are well-known technical means. This application will not elaborate further on them here.

[0031] S2: Based on the child's real-time state vector and the interaction event type label, simultaneously read the child's growth profile and the robot's life form profile. Using this interaction event as a basis, perform synchronous incremental updates and output the updated child's growth profile and robot's life form profile.

[0032] For the current interaction event type label, based on keyword matching of the voice content and the context of the interaction scenario, the interaction content is categorized to determine the interaction event type label. Types include, but are not limited to: English learning events, math training events, programming enlightenment events, reading companionship events, movement imitation events, scientific observation events, free dialogue events, and rest events. Rest events refer to all event types other than the first seven. The type label is output by the system's built-in classifier, with a confidence threshold set above 0.75. The label's role is to determine which one or more dimensions of the child's developmental profile are updated in this interaction, and simultaneously determine the target skill branch for the coupled mapping operation.

[0033] When the robot first interacts with a child, the operator creates a preliminary profile of the child based on the purchaser's description, which serves as the initial data for the child's growth profile. The robot's life form profile is generated after at least one interaction with the child, based on the preliminary profile of the child and the interaction record vector. The child's growth profile and the robot's life form profile are dynamically updated based on subsequent interaction records.

[0034] In this embodiment, the child's growth profile is updated. The child's growth profile adopts a multi-dimensional vector structure and has several independent dimensions, such as seven independent dimensions (corresponding to seven learning cases), including English proficiency, mathematical thinking, reading comprehension, programming logic, motor coordination, creative expression, and concentration habits. The score range for each dimension is [0, 100].

[0035] Based on the event type tag, determine the target dimensions that need to be updated: if the event type is an English learning event, update the English proficiency level; if it is a math training event, update the math dimension, and so on. For the target dimensions, based on the focus score and interaction initiative score, combined with the duration of this interaction (in minutes), calculate the growth increment and update the target dimensions of the child's growth profile accordingly.

[0036] The calculation of this growth increment uses effective learning time as an intermediate variable. The effective learning time is calculated as follows: ;in: This indicates the effective learning time, expressed in minutes. Indicates the duration of this interaction, in minutes; This represents the focus score, ranging from 0 to 100. This represents the initiative coefficient. The original interaction time is discounted by both the child's level of focus and initiative, and then converted into equivalent learning time.

[0037] The piecewise function of the initiative coefficient is: ;in: Indicates the initiative coefficient; The score represents the initiative of interaction, ranging from 0 to 100; children's proactive interactive behaviors have a higher learning conversion efficiency than passively responding interactive behaviors.

[0038] Single-time limit truncation: ;in: This indicates the effective learning time after truncation, in minutes. Indicates effective learning time; This represents the maximum effective learning time limit for a single session. It is a system-preset constant with a value range of 30-60 minutes and can be dynamically set. There is a physiological upper limit to the effective learning time for a single interaction. The marginal contribution of the portion exceeding this limit to cognitive gain approaches zero.

[0039] The original value of the dimension score increment is: ;in: Represents the original value of the uncapped dimension score increment; This indicates the effective learning time after truncation; This represents the preset difficulty weight coefficient for dimension d, with values ​​ranging from 0.2 to 0.8 for each dimension. It represents the percentage increase in scores that can be converted from each minute of effective learning time. When it equals 0.3, it means that each minute of effective learning can generate a mastery increase of 0.3 points. The unit is increase in points per minute. Different learning domains have different cognitive loads and growth rates, and the difficulty weight coefficient is used to differentiate and calibrate the single growth rate of different dimensions.

[0040] The specific preset difficulty weight coefficients for each dimension at the time of manufacture are as follows: English proficiency 0.65, reading comprehension 0.60, motor coordination 0.70, concentration habit 0.75, mathematical thinking 0.45, creative expression 0.40, and programming logic 0.28. These values ​​are fixed at the time of manufacture and will be fine-tuned within ±0.05 based on the child's actual learning performance.

[0041] The upper limit for a single increment is truncated as follows: ;in: This represents the increase in growth, i.e., the final increase in the dimension score; This represents the original value of the dimension score increment; This represents the maximum allowable increment in a single interaction, which is a system-preset constant with a value of 10 points. Even if a child performs exceptionally well, there is a physiological limit to the cognitive improvement after a single interaction, and increments exceeding 10 points are not statistically reliable.

[0042] Dimension scores updated to: in: This represents the mastery score for dimension d after the update. This represents the mastery score of dimension d before the update, with a value ranging from 0 to 100; This represents the final dimension score increment; the mastery score has clear upper and lower semantic boundaries, with 0 indicating no mastery and 100 indicating complete mastery, and any calculated value exceeding this boundary has no practical significance.

[0043] The execution of all the above calculations must meet the following prerequisites: ;in The minimum effective interaction time threshold preset for the system is set to 30 seconds. If this occurs, the interaction is considered invalid. The score is set to 0 directly, without triggering any score updates. Interactions that are too short to form effective learning behaviors are not included in the growth profile statistics.

[0044] For the other six dimensions not covered in this update, a natural forgetting mechanism is introduced: 2% of the current value of each unupdated dimension is subtracted as the forgetting decay, with a minimum decay of 80% of the initial baseline value of the dimension. Thus, the child's growth profile is updated for this time. This step outputs a seven-dimensional increment vector composed of the actual changes in each dimension, with positive values ​​representing growth and negative values ​​representing forgetting.

[0045] In this embodiment, the robot life form profile is updated. The robot life form profile adopts an independent data structure and records the following fields: current level (positive integer, initially 1), current experience value (integer, initially 0), current mood value (0 to 100), current intimacy level (0 to 100), current energy value (0 to 100), skill tree unlock status (lock or unlock markers for each skill node corresponding to different learning events), learned action library index list, learned voice library index list, learned lighting effect mode index list, and plot progress (integer, recording the number of completed nodes).

[0046] Based on the interaction event type label, attention score, and emotion polarity score, the experience increment in the robot life form profile is calculated.

[0047] The starting point is the base experience value determined by the event type. The base experience values ​​are as follows: English event 80, Math event 100, Programming event 120, Reading event 90, Sports event 70, Scientific observation event 85, Free conversation event 40, and Rest event 20.

[0048] The focus gain coefficient is calculated as follows: When the focus score is 50, the coefficient is 1, meaning it is neither amplified nor reduced; when the focus score is below 50, the coefficient is less than 1, and the experience value is reduced; when the focus score is above 50, the coefficient is greater than 1 but not exceeding 1.5, and the experience value is appropriately rewarded.

[0049] The emotion correction coefficient is set as follows: 1.1 when the emotion polarity score is positive, 1.0 when it is zero, and 0.9 when it is negative. Positive emotions amplify the acquisition of experience points, negative emotions weaken the acquisition of experience points, and neutral emotions maintain their original value, so that the robot's growth rate is correlated with the emotional atmosphere during the interaction.

[0050] This experience increment is defined as: ;in, This is the increase in experience points for this instance; These are basic experience points; It is the focus gain coefficient; This is the emotion correction coefficient. First, the base experience value is multiplied by the focus gain coefficient and the emotion correction coefficient to get the product. Then, it is rounded to the nearest integer and finally compared with 200 to take the smaller value, ensuring that the single increment does not exceed the upper limit of 200.

[0051] The experience increment is added to the current experience value. Then, it is determined whether the current experience value has reached the upgrade threshold. The threshold is calculated as 5 times the square of the level. If it is reached, the level increases by 1 and the experience value is reset to zero and accumulated again. At the same time, the level upgrade event is recorded.

[0052] Let the robot's current experience value before the interaction be... If the current level is L (a positive integer greater than 1), then the experience points after the interaction will be temporarily recorded as: The upgrade threshold corresponding to the current level is defined as: The threshold increases quadratically with the level, ensuring that higher levels require more experience to upgrade, which aligns with the long-term growth curve.

[0053] like This will trigger an upgrade, and the level and experience points will be updated as follows: ;in, This refers to the experience points accumulated after leveling up to the next level (i.e., the original experience points minus the level-up threshold).

[0054] like If not, then do not upgrade: A single interaction may trigger multiple upgrades due to a large increase in experience.

[0055] Let the remaining experience value after the upgrade be... The current level is The condition for continuous upgrades can be recursively expressed as: when hour: ;in, This indicates an assignment operation, where the value calculated on the right side is used to update the variable on the left side. Repeat the above judgment and assignment until... until.

[0056] final This is the final level after this update. This is the accumulated experience points for the next level. Each time an upgrade is triggered, a level increase event is recorded.

[0057] Mood values ​​are updated based on the emotional polarity score and the amount of experience gained: for every 10 points increase in emotional polarity score, mood values ​​increase by 3 points; for every 50 points of experience gained, mood values ​​increase by 1 point. A daily natural decay mechanism is also introduced: mood values ​​decrease by 2 points for every 2 consecutive hours of inactivity, but the upper limit of mood values ​​is 100 and the lower limit is 0.

[0058] The intimacy level is updated solely based on the total number and duration of touch events recorded by the touch sensor during the current interaction: each touch event adds 0.5 points, and every 5 seconds of continuous touch adds an additional 1 point, but the maximum increase per interaction is 10 points. A decay rule is also set where intimacy decreases by 5 points daily with each consecutive day of no interaction.

[0059] The energy value is updated based on the duration of the interaction: the energy value decreases by 3 points every 10 minutes of interaction. When the energy value is below 20 points, the robot automatically enters a low-power standby state and rejects new learning-type interaction requests, only responding to charging or rest commands.

[0060] S3: Perform bidirectional coupling mapping calculations based on the updated child growth profile and robot life form profile, and output skill node unlock instruction set; use the unlock instruction set to update the robot life form state model.

[0061] The bidirectional coupling mapping operation is set as follows: read all preset skill nodes in the robot skill tree. Each skill node is pre-bound with a unique unlock condition tuple at the factory. The tuple consists of the required child mastery threshold and the required robot level threshold.

[0062] The required mastery threshold for children directly points to a specific dimension in the child's development profile. Skill nodes correspond one-to-one with dimensions. For example, the skill "encouragement of active English dialogue" is essentially an improvement in the robot's expressive ability in English interaction scenarios. Therefore, its mastery threshold points to the "English mastery" dimension in the child's development profile. The skill "guidance on programming action arrangement" points to the "programming logic" dimension.

[0063] The required robot level threshold is a positive integer, with a value corresponding to the depth of the skill node in the skill tree. The skill tree is organized in a hierarchical structure: the first layer consists of basic skills (such as basic encouraging voices and simple actions), with a required level threshold of 1 or 2; the second layer consists of advanced skills (such as domain-specific dialogues and combined actions), with a required level threshold of 3 to 5; and the third layer consists of high-level comprehensive skills (such as cross-domain guidance and creative task arrangement), with a required level threshold of 6 to 10. The level in the robot's life form profile reflects the total amount of interaction accumulated between the child and the robot over a long period. The higher the level, the more interactions the child has completed and the more basic skills the robot has unlocked. Tying higher level thresholds to higher-level skills ensures that children have experienced a sufficient number of basic interactions and accumulated the necessary prior knowledge before attempting to unlock higher-level skills.

[0064] The current absolute values ​​of all dimensions in the updated child development profile are compared with the required child mastery threshold for each unlocked skill node. Simultaneously, the current level in the updated robot lifeform profile is compared with the required robot level threshold for the same skill node. Only when both of these conditions are met is the corresponding skill node marked as unlockable for this time.

[0065] For example, if the unlocking conditions for the robot's "English Active Conversation Encouragement" skill node are "child's English proficiency score is greater than or equal to 80 points" and "robot level is greater than or equal to 5," then the skill node will only be marked as unlockable when the child achieves an English proficiency score of 80 points or higher through genuine learning behavior and the robot's current level is at least 5. The system does not have any operational path that can bypass the child's actual learning behavior to directly unlock the skill.

[0066] For each skill node marked "Unlockable This Time", further query the corresponding embodied expression resources associated with the node, including a voice text template, a set of servo action parameter sequences, a combination of lighting effect color and breathing frequency, and a projection animation identifier; write the index number of the embodied expression resource along with the skill node ID into the skill node unlock command set. If no new skill node meets the unlock conditions this time, the command set is empty.

[0067] The update of the robot life form state model follows the execution of the coupling mapping operation. The robot life form state model is a persistent data structure stored in the robot's non-volatile memory, existing alongside the robot life form profile. Internally, it is divided into two hierarchical regions: the causal response layer and the derived statistics layer, with completely different data update mechanisms and triggering conditions for the two layers.

[0068] The causal response layer records the embodied changes directly resulting from the changes in the child's growth profile. This layer includes a group of ability fields, a group of growth scale fields, and a group of narrative progress fields.

[0069] The capability field group includes a skill node status table, a learned action library index table, a learned voice library index table, and a learned lighting effect mode index table. The skill node status table stores the current status of each of all preset skill nodes, which is either "locked" or "unlocked". When the skill node unlock instruction set output by this coupling mapping operation is not empty, the system modifies the corresponding status of all skill nodes in the instruction set in the skill node status table to "unlocked", and appends the associated action resource number to the learned action library index table, the voice resource number to the learned voice library index table, and the lighting effect resource number to the learned lighting effect mode index table.

[0070] The growth scale field group includes two dimensions: level and growth stage. The positive integer value of the level is synchronized with the level field in the robot life form profile, and this field is updated synchronously each time the level in the profile increases. The growth stage is an enumeration type, with selectable values ​​of "Early Childhood Companionship Period," "Adolescent Co-learning Period," "Youth Mentorship Period," and "Creative Leadership Period," with the initial value being "Early Childhood Companionship Period." When the level reaches the preset stage jump threshold (Level 5, Level 12, Level 25) and the absolute value of mastery in at least three dimensions in the child's growth profile simultaneously reaches the preset threshold, this field automatically changes to the next stage. After the growth stage changes, the system automatically loads a new personality parameter configuration set that matches the new stage.

[0071] The narrative progress field group contains two dimensions: story progress and world-building nodes. Story progress is an integer variable that records the sequence number of completed story nodes in the current plot sequence, with an initial value of 0. The story progress increments by 1 whenever the cumulative total increment of all seven dimensions in the child's growth profile reaches a set threshold of 50 points. World-building nodes are string identifiers that point to the projection scene resource to be displayed. Their values ​​are linked to the story progress; for example, story progress of 1 to 3 corresponds to "Language Port," 4 to 6 to "Logic Valley," 7 to 9 to "Creation Workshop," and 10 and above to "Observation Starfield."

[0072] The derived statistics layer does not directly respond to the current coupling mapping operation. Instead, it uses the historical sequence of the robot's life form profile from the last 30 interactions as its data source to perform periodic batch derivation calculations, including growth rate, growth stability, emotion recovery half-life, interaction type preference vector, and effective arousal period. The derived statistics layer refreshes every 10 interactions. During this execution, it first determines whether 10 interactions have been accumulated since the last refresh: if not, it skips this layer's update; if so, it reads the profile history of the last 30 interactions and performs all the following derivation calculations: The growth rate is equal to the net increment of the level field in the most recent 30 interactions divided by the actual number of calendar days spanned by those 30 interactions, expressed in levels per week; it is used to reflect the speed trend of progress of children driving robots.

[0073] Growth stability is equal to the standard deviation of the rating values ​​over the last 30 interactions. A smaller standard deviation indicates more stable growth and is used to assess whether a child's interaction habits are regular.

[0074] The emotional recovery half-life refers to the average number of interactions required for the mood value in the image to recover to the median value after it has decreased due to the child's negative emotional feedback; it is measured in cycles and reflects the robot's ability to recover its emotions after being affected by the child's emotions.

[0075] The interaction type preference vector is a statistical distribution of the frequency of eight event types in the most recent 30 interactions. It is stored in the form of an eight-dimensional vector, with the value of each dimension representing the percentage of the corresponding type's occurrences out of the total number of interactions.

[0076] Effective wake-up periods were obtained by statistically analyzing the hourly density of timestamps of the most recent 30 interactions initiated by the child, and marked as the two consecutive hourly periods with the highest frequency of occurrence.

[0077] In this embodiment, a direct causal connection is established between a child's growth profile and the robot's embodied abilities, so that the child's improvement in a certain area can trigger the unlocking of new robot skills. At the same time, the robot's life form state model is used to convert single unlocking events into long-term behavioral patterns.

[0078] S4: Receives the child's real-time state vector, the updated robot life form profile, the derived statistics layer fields in the updated robot life form state model, and a set of configurable personality parameters.

[0079] The personality parameters include five dimensions: gentleness, principledness, mischievousness, curiosity, and companionship. Each parameter has a value range of 0 to 100 and is set to the default value at the factory.

[0080] The personality arbitrator first integrates the above inputs into decision-making basis data, and then executes three layers of decision-making logic in sequence.

[0081] The first layer is the interaction intent decision. Attention score, fatigue score, and emotional polarity score are combined into a comprehensive demand index. The comprehensive demand index is calculated as follows: ;in, It is a comprehensive demand index; It is a focus score; It is the fatigue score, calculated from the above content, with a value ranging from 0 to 100; It is an emotion polarity mapping score. The emotion polarity score is obtained through linear mapping, and the mapping formula is: Where E is the original emotional polarity score.

[0082] The emotional polarity mapping score ranges from 0 to 100, with higher values ​​indicating a more positive and active emotional state for the child. In the formula, the weight is 0.3, which is used to balance the influence of emotions on the willingness to interact. When emotions are positive, it is easier to trigger proactive interaction, while when emotions are negative, there is a tendency to reduce forced interaction.

[0083] A higher fatigue score indicates that the child's current energy level is lower and they are closer to needing rest, thus reducing the overall demand index and avoiding forcing interaction when the child is fatigued. The fatigue score also has a weight of 0.3, the same as the emotion weight, reflecting equal importance placed on the child's physiological state.

[0084] Simultaneously, the effective wake-up period in the derived statistics layer of the robot life form state model is read to determine whether the current moment is within the effective wake-up period.

[0085] If the comprehensive demand index is greater than 55, the current time since the last interaction has exceeded the preset minimum interval of 90 seconds, and the current time is within the effective wake-up period or the current time period has been marked as allowed by the parent rules, then the arbitrator will output the intent to "actively initiate interaction".

[0086] If the fatigue score is greater than 70 and the current interaction duration has exceeded 15 minutes, output the "Recommend Rest" intent.

[0087] If the overall demand index is between 30 and 55, then the intention to "silently accompany" will be output.

[0088] If the overall demand index is below 30, the message "Do not disturb" will be displayed.

[0089] The second layer is for interaction strategy selection, which is executed only when the output of the first layer is "actively initiate interaction". The arbitrator identifies the three dimensions with the lowest current percentile based on the long-term percentile ranking of each dimension in the child's growth profile. At the same time, it reads the interaction type preference vector from the derived statistics layer of the robot life form model and identifies the two most frequent interaction types as preference types.

[0090] If the initiative score is below 40, the strategy is to encourage, such as using positive affirmation and mild challenges, with a gentle tone and low information density; if the initiative score is between 40 and 70, the strategy is to guide, such as using demonstration and questioning, with a steady tone and medium information density; if the initiative score is above 70, the strategy is to explore, such as using open-ended creative tasks, with a lively tone and high information density.

[0091] Meanwhile, the arbitrator reads the emotion recovery half-life in the derived statistics layer of the robot's life form state model: if the emotion recovery half-life is greater than 5 interactions, it means that the robot's emotion recovery is slow. The arbitrator automatically shortens the expected duration of this interaction to 70% of the default duration to avoid the robot's emotion from further declining due to excessive interaction.

[0092] The third layer is for determining the intensity of emotion expression, which is executed under all intention types. The arbitrator jointly determines the dominant emotion type and intensity of the robot's interaction based on the emotion polarity score and the current mood value in the robot's life form profile. The dominant emotion type is selected from six basic emotions in the robot's built-in emotion library: happiness, anticipation, calmness, disappointment, coquetry, and slight anger.

[0093] The selection rules are as follows: if the arithmetic mean of the emotion polarity score and mood value is higher than 60, the person tends to be happy or expectant; if the mean is between 35 and 60, the person tends to be calm or coquettish; and if the mean is lower than 35, the person tends to be disappointed or slightly angry. The maximum permissible expression intensity of the latter two is suppressed by the weighted index of gentleness in the personality parameter. For every 10-point increase in gentleness, the maximum permissible expression intensity of disappointment and slightly angry decreases linearly by 5%.

[0094] The intensity of expression is ultimately output as a value from 0 to 100, and is modulated by the growth stability in the derived statistical layer of the robot life form state model: the higher the growth stability, the smaller the successive fluctuation of the intensity of expression, ensuring the continuity of emotional output.

[0095] This embodiment integrates multi-source heterogeneous data from the perception layer, image layer, state model layer, and long-term statistics layer into a multi-dimensional decision-making framework. This ensures that each interaction decision of the robot takes into account the child's current state, long-term growth needs, the robot's own attributes, and historical behavior patterns, thus avoiding inconsistencies or rigidity in the robot's behavior.

[0096] S5: Receive the skill node unlock instruction set, interaction intent, strategy type, and emotional expression intensity, and execute the three sub-tasks of skill unlock execution, achievement judgment, and interaction content synthesis in sequence to generate the final interaction content and instruction operation.

[0097] The first subtask is skill unlocking. It checks if the skill node unlocking instruction set is empty. If not, it submits the associated resource index numbers of all skill nodes in the instruction set to the resource scheduler. The resource scheduler loads the corresponding voice text template into the voice synthesis queue, the servo action parameter sequence into the motion control buffer, the lighting effect parameters into the LED driver register, and the projection animation identifier into the projection content cache. After all loading is complete, it internally marks the interaction as having "growth display content." If the instruction set is empty, it skips the loading process and marks the interaction as not having growth display content.

[0098] The second sub-task is achievement determination. Based on the incremental vectors of each dimension of the child's growth profile and the growth scale fields of the causal response layer of the robot life form state model, it is determined whether a preset achievement has been triggered. Each preset achievement is bound to a trigger condition tuple at the factory: the minimum number of consecutive days required to achieve the goal and the minimum net growth in the required dimension. The minimum number of consecutive days required to achieve the goal refers to the minimum number of days the child has achieved positive growth in a specific dimension, and the minimum net growth in the required dimension refers to the cumulative net growth value of that dimension from the first date of achievement to the current day.

[0099] The current absolute value of the target dimension for this update is compared with the archived value of the corresponding dimension 30 days ago to calculate the net growth. If the net growth reaches a preset threshold for a certain achievement, and the number of consecutive days with positive growth in the corresponding dimension reaches the preset minimum number of days for the corresponding achievement, then the corresponding achievement is marked as "triggered". After the achievement is triggered, the badge data corresponding to the achievement is written to the permanent record area of ​​the robot's life form portrait, and an achievement congratulatory message is generated simultaneously, including a specific congratulatory voice and a set of celebratory action parameters, and added to the output queue of this interaction. The determination of the achievement does not depend on a single score or on-the-spot performance, but on whether the net growth of a specific dimension in the child's growth portrait accumulated over days reaches a preset milestone threshold.

[0100] The third subtask is the synthesis of embodied interactive content. Basic interactive content is organized based on the type of the output interaction intent: if the intent is "initiating interaction proactively," then voice text is selected from the corresponding corpus according to the strategy type. For example, an encouragement strategy selects the text from the encouragement corpus that best matches the weak dimension information, accompanied by a set of small nodding and tail-wagging movements and warm-colored breathing light effects; a guidance strategy selects a heuristic question text related to the preference type from the guidance corpus, accompanied by a slight head tilt and cool-colored gradient light effects; an exploration strategy selects an open-ended challenge task description from the creative task library, accompanied by a rotation movement and multi-colored flashing light effects.

[0101] If the intention is to "suggest a rest", then play a rest reminder voice message accompanied by a head-lowering gesture and warm-colored constant lighting effects.

[0102] If the intention is "silent companionship" or "non-disturbing", then only a very low brightness light effect and slow breathing are used as a passive presence indication, without outputting voice or movement content.

[0103] Then, the growth display content (if any), achievement celebration content (if any), and the aforementioned strategy content are sequentially merged. The merging rules are as follows: achievement celebration content has the highest priority and, if present, is placed at the beginning of the interactive content sequence; growth display content is next, placed after achievement celebration content and before strategy content. The total duration of the overall presentation is limited to within 15 seconds.

[0104] The merged content sequence is output synchronously through each execution channel. The speech text is converted into a speech signal with emotion parameters by the local TTS engine and played through a speaker. These emotion parameters, controlled by the dominant emotion type and intensity, regulate speech rate, pitch, and pause rhythm. Servo motion parameters are calculated using inverse kinematics to drive joint motors. Lighting parameters drive an RGB-LED array to produce corresponding colors and breathing patterns. The projected content is projected onto an interactive surface in front of the child via a DLP optical engine. The output timing of all channels is synchronized by the system clock, ensuring that voice broadcasting, motion execution, lighting changes, and projection display are presented collaboratively within the same time window.

[0105] The abstract decision data generated in the preceding steps are transformed into multi-dimensional physical expressions that children can directly perceive, so that the robot's response not only reflects personality consistency but also presents visible, audible, and perceptible embodied changes. At the same time, the achievement system solidifies children's long-term growth milestones in a recordable and displayable form.

[0106] S6: After each interactive output is completed, collect the child's feedback data and re-execute steps S1-S5 for real-time updates.

[0107] In some embodiments, within a 15-second window following the completion of the interactive content output, the child's subsequent reactions are continuously monitored. Based on the data collection operations described above, a feedback dataset (consistent with the original multidimensional data content) is obtained. The four data points are weighted and summed to form a comprehensive feedback score, normalized to a range of -100 to +100. This score is appended to the interaction record and stored in the historical archive along with the profile update data for this interaction.

[0108] Check the cumulative number of interactions since the last refresh of the derived statistics layer. If it has reached 10 interactions, perform a batch refresh of the derived statistics layer: read the complete record of the last 30 interactions from the historical archive, recalculate all five indicators—growth rate, growth stability, emotional recovery half-life, interaction type preference vector, and effective arousal period—according to the defined derived calculation method, and overwrite the corresponding fields in the model with the calculation results. If it has less than 10 interactions, skip the refresh this time, and retain the existing derived statistics values.

[0109] Simultaneously, based on the average comprehensive feedback score of the most recent 20 interactions, the personality parameters of gentleness, principledness, mischievousness, and curiosity are fine-tuned within a range of no more than plus or minus 3 points: if the average feedback score is greater than +30, mischievousness and curiosity are increased by 2 points each; if the average feedback score is less than -30, gentleness and principledness are increased by 2 points each. The fine-tuned personality parameters are written back to the system configuration file for use in subsequent interaction cycles.

[0110] The current absolute values ​​of each dimension in the child's developmental profile are compared with the historical archive values ​​from 90 days ago to calculate the net growth in each dimension. If the net growth in at least three of the following dimensions—English proficiency, mathematical thinking, reading comprehension, programming logic, and creative expression—exceeds 20 points, and the current level in the robot life form profile is not lower than level 10, and the cumulative effective companionship days are not less than 180 days, then the conditions for a personality stage leap are met.

[0111] This approach enables the entire method to learn online and evolve over a long period. Interactive feedback is used to optimize short-term personality performance, periodic updates ensure that the state model keeps up with changes in children's behavioral patterns, and long-term accumulated growth data drives the robot to undergo phased personality leaps, achieving a continuous positive cycle between children's growth and robot evolution.

[0112] In this embodiment, the target dimensions in the child's growth profile are incrementally updated through the profile update step, while the robot's life form profile is updated synchronously. This allows the child's learning progress and the robot's state changes to be recorded bidirectionally within the same interaction cycle. This solves the technical problem in the prior art that children's learning outcomes cannot be structured and stored to drive changes in the robot, and achieves the technical effect of dual-track traceable recording of children's cognitive growth and robot state changes.

[0113] By using a coupling mapping step, the current mastery level of each dimension of the child's growth profile is compared with the mastery threshold of preset skill nodes. At the same time, the robot's current level is compared with the level threshold of the skill node. Only when both conditions are met simultaneously, the corresponding skill node is unlocked and the causal response layer of the robot's life form state model is updated. Meanwhile, the derived statistics layer is refreshed periodically. This ensures that the unlocking of new robot skills strictly depends on the simultaneous fulfillment of two independent conditions: the child's actual ability in the corresponding learning area and the robot's long-term accumulated level. This achieves the technical effect of making the child's real learning behavior the sole driving path for the robot's embodied ability improvement, fundamentally eliminating the possibility of acquiring ability growth through non-learning behaviors.

[0114] Through the comprehensive demand index calculation and three-layer decision-making mechanism of the personality arbitration process, the interaction intention decision, interaction strategy selection, and emotional expression intensity are all adaptively generated based on the child's immediate state, dual-profile data, and derived statistical layer data. This achieves the technical effect of the robot interacting with the child at the right time, with the right strategy, and with the right emotional expression. This constitutes a complete closed loop from perception to decision-making to expression, enabling the robot to adaptively generate personalized interactive responses based on the child's real learning events.

[0115] Example 2: This example also provides an AI personality generation system for children's adaptive companionship interaction, such as... Figure 2 As shown, the system includes: The perception and recognition module is used to synchronously collect multi-dimensional raw signals from children within a preset sampling period to form an interaction record vector. Based on the interaction record vector, time window aggregation and feature calculation are performed to obtain the child's instantaneous state vector and the current interaction event type label. The multi-dimensional raw data includes a sequence of frontal facial images, three-dimensional spatial coordinates, voice signals, touch event data, relative motion data, and timestamps. The child's instantaneous state vector is a sliding window formed by taking all interaction record vectors within the previous 30 seconds as the endpoint, and calculating the child's state indicators. The state indicators include focus score, fatigue score, emotional polarity score, interaction initiative score, and physical activity intensity score.

[0116] The profile update module receives the child's real-time state vector and interaction event type label from the perception and recognition module. Based on the child's real-time state vector and interaction event type label, and taking the current interaction event as the basis, it performs synchronous incremental updates and outputs the updated child growth profile and robot life form profile.

[0117] The coupling mapping module is used to receive the updated child growth profile and robot life form profile output by the profile update module; read all preset skill nodes in the robot skill tree; perform bidirectional coupling mapping operation based on the updated child growth profile and robot life form profile; output skill node unlocking instruction set; and update the robot life form state model using the unlocking instruction set; the robot life form state model includes a causal response layer and a derived statistics layer.

[0118] The personality arbitration module receives the child's real-time state vector output by the perception and recognition module, the updated robot life form portrait output by the portrait update module, the derived statistical layer fields in the updated robot life form state model output by the coupling mapping module, and a set of configurable personality parameters; it executes the decision-making mechanism to obtain the interaction intent, interaction strategy, and emotional expression intensity.

[0119] The interaction generation module receives the skill node unlocking instruction set output by the coupling mapping module and the interaction intent, strategy type, and emotional expression intensity output by the personality arbitration module; it then executes three sub-tasks in sequence: skill unlocking execution, achievement judgment, and interaction content synthesis to generate the final interaction content and instruction operation.

[0120] In this system, the perception and recognition module transforms the naturally occurring, unstructured, multimodal interactions between children and robots into numerical data elements that can be quantitatively processed by subsequent computing units. Its technical effect is to enable subsequent profile updates and personality decisions to be based on a high-level understanding of the child's state rather than low-level physical signals, thus solving the problem that raw perception data cannot be directly used for cognitive state inference.

[0121] The output of the perception and recognition module is connected to the input of the profile update module and the personality arbitration module. The five-dimensional real-time state vector of the child calculated by the perception and recognition module is supplied to the profile update module as the basis for calculating the growth increment, and to the personality arbitration module as the input for calculating the comprehensive demand index; the current interaction event type label output by the perception and recognition module is supplied separately to the profile update module to determine the target dimension of the child's growth profile for this update.

[0122] The profile update module transforms the child's performance in this interaction into two different types of quantitative updates. The technical effect is that the child's growth profile reflects the cumulative changes in the child's long-term abilities, while the robot life form profile reflects the dynamic attributes of the robot as a companion. The two provide a dual data foundation for subsequent coupling mapping and personality decision-making, solving the problem that a single profile cannot simultaneously represent the child's cognitive development and the robot's state changes.

[0123] The input of the portrait update module is connected to the output of the perception and recognition module, and its output is connected to the input of the coupling mapping module and the personality arbitration module. The updated child growth portrait and robot life form portrait output by the portrait update module together serve as the data source for skill node unlocking judgment in the coupling mapping module; the updated current value of the robot life form portrait is also separately supplied to the personality arbitration module as input for mood value-related decisions.

[0124] The role of the coupling mapping module is to establish a direct causal connection between a child's growth profile and the robot's embodied capabilities. Its technical effect is that a child's improvement in a certain area can trigger the unlocking of new robot skills. At the same time, through the derived statistical layer of the robot's life form state model, the single unlocking event is precipitated into a long-term behavioral pattern, which solves the problem of the lack of quantitative mapping relationship between children's learning progress and changes in robot capabilities.

[0125] The input of the coupling mapping module is connected to the output of the profile update module, and its output is connected to the input of the personality arbitration module and the interaction generation module. The skill node unlocking instruction set output by the coupling mapping module is directly supplied to the interaction generation module for resource loading of skill unlocking execution sub-tasks; the derived statistical layer fields in the robot life form state model output by the coupling mapping module are supplied to the personality arbitration module as auxiliary decision-making basis for effective arousal period determination, strategy selection, and emotion intensity modulation.

[0126] The role of the personality arbitration module is to integrate multi-source heterogeneous data from the perception layer, profiling layer, state model layer and long-term statistical layer into a multi-dimensional decision-making framework. Its technical effect is to enable the robot to take into account the child's current state, long-term growth needs, robot's own attributes and historical behavior patterns in every interaction decision, thus solving the problem of inconsistencies or rigidity in robot behavior.

[0127] The input of the personality arbitration module is simultaneously connected to the outputs of the perception recognition module, the profile update module, and the coupling mapping module, while its output is connected to the input of the interaction generation module. The three decision results output by the personality arbitration module—interaction intention, strategy type, and emotional expression intensity—serve as control parameters for the synthesis of interaction content in the interaction generation module.

[0128] The interactive generation module transforms the abstract decision data generated by the preceding modules into a multi-dimensional physical expression that children can directly perceive. Its technical effect is to make the robot's response reflect both personality consistency and tangible changes that are visible, audible, and perceptible. At the same time, the achievement system solidifies children's long-term growth milestones in a recordable and displayable form, solving the problem that abstract decision data cannot be directly perceived and understood by children.

[0129] The input end of the interaction generation module is connected to the output end of the coupling mapping module and the personality arbitration module. Its output end generates the final interaction content and instruction operation. After execution, the child's immediate reaction to the robot's feedback during this interaction is re-collected and recorded as input data for subsequent interaction cycles, forming a closed-loop feedback.

[0130] The entire system achieves a complete adaptive companion interaction closed loop from perception to decision-making to expression through the sequential execution and data transmission of the above five modules. This enables the robot to update its own life status in real time and generate new interactive capabilities based on the child's learning events.

[0131] To make the technical solution of the present invention clearer, the above specific embodiments may involve several specific numerical values, thresholds, parameters, formulas, and examples. Those skilled in the art should understand that these specific contents are merely exemplary descriptions given for ease of implementation and are not essential technical features necessary for realizing the present invention, nor do they constitute a limitation on the scope of protection of the present invention. The mathematical formulas, parameter values, model structures, calculation rules, constraints, and data acquisition methods described in the specific embodiments section of this application are preferred examples for realizing the technical concept of the present invention, not the only implementation methods. Based on the core logic and inventive concept disclosed in this application, those skilled in the art can make reasonable equivalent substitutions, parameter adjustments, or logical optimizations to the parameters, formulas, and algorithm models according to actual conditions, industry standards, or conventional experimental methods, without producing substantial differences. Any changes that do not depart from the inventive concept of this application fall within the scope of this application. All mathematical expressions in this application should ensure that their physical meaning is clear when applied; parameters such as weights and thresholds can be calibrated using statistical or machine learning methods known in the art.

[0132] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating an AI personality for adaptive companionship interaction with children, characterized in that, include: S1: Collect multidimensional raw data of children to form an interaction record vector. Based on the interaction record vector, perform time window aggregation and feature calculation to obtain the child's real-time state vector and the current interaction event type label. S2: Based on the child's real-time state vector and the interaction event type label, and taking this interaction event as the basis, perform synchronous incremental updates and output the updated child growth profile and robot life form profile; S3: Perform bidirectional coupling mapping calculations based on the updated child growth profile and robot life form profile, and output a skill node unlocking instruction set; update the robot life form state model using the unlocking instruction set; the robot life form state model includes a causal response layer and a derived statistics layer. S4: Based on the child's immediate state vector, the updated robot life form profile, the derived statistical layer fields and personality parameters in the updated robot life form state model, execute the decision-making mechanism to obtain the interaction intention, interaction strategy and emotional expression intensity. S5: Receive the skill node unlock instruction set, interaction intent, strategy type and emotional expression intensity, and execute the three sub-tasks of skill unlock execution, achievement judgment and interaction content synthesis in sequence to generate the final interaction content and instruction operation; The multidimensional raw data includes a sequence of frontal facial images, three-dimensional spatial coordinates, voice signals, touch event data, relative motion data, and timestamps; The child's real-time state vector is a sliding window formed by taking all interaction record vectors within the previous 30 seconds as the endpoint, and calculating the child's state index; the state index includes focus score, fatigue score, emotional polarity score, interaction initiative score and physical activity intensity score.

2. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 1, characterized in that, In step S1, a sequence of frontal facial images of the child is acquired using a visual camera, including: extracting the coordinates of facial key points, calculating the child's current head yaw angle and pitch angle; and simultaneously outputting the facial motion unit intensity value. The three-dimensional spatial coordinates of the center point of the child's chest relative to the robot's body coordinate system are obtained by using a DTOF sensor, and the straight-line distance between the child and the robot is calculated. Children's speech signals are collected using a microphone array, and speech recognition and detection are performed to detect valid text and extract the average energy value and speech rate estimate of the speech segment. By detecting the robot's own posture angle and motion state using an IMU, and combining this with the rate of change of the child's relative position, the child's movement speed and activity intensity in the interactive space can be estimated. The touch event data includes the touch event type and duration, the output touch force level, and the duration of a single touch.

3. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 1, characterized in that, In step S2, the interactive content is categorized based on keyword matching of the voice content and the context of the interaction scenario, and the interaction event type label is determined. The types include: English learning event, math training event, programming enlightenment event, reading companionship event, sports imitation event, scientific observation event, free dialogue event, and rest event.

4. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 1, characterized in that, Perform synchronous incremental updates, including: for the child's growth profile, determine the target dimension for this update based on the type label of this interaction event; for the target dimension, calculate the growth increment and update the target dimension of the child's growth profile based on the focus score and interaction initiative score, combined with the duration of this interaction; for other dimensions not involved in this update, introduce a natural forgetting mechanism: for each unupdated dimension, subtract 2% of its current value as the forgetting decay amount, with a minimum decay of 80% of the initial baseline value of the dimension.

5. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 4, characterized in that, Performing synchronous incremental updates also includes: for the robot life form profile, calculating the current experience increment in the robot life form profile based on the current interaction event type label, attention score, and emotion polarity score; including: starting from the basic experience value determined by the event type, calculating the attention gain coefficient and emotion correction coefficient to obtain the current experience increment; adding the experience increment to the current experience value, determining whether the current experience value has reached the upgrade threshold, if it has, increasing the level by 1 and resetting the experience value to zero to start accumulating again, while recording the level upgrade event.

6. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 1, characterized in that, The bidirectional coupling mapping operation is set as follows: read all preset skill nodes in the robot skill tree. Each skill node is pre-bound with a unique unlocking condition tuple when it leaves the factory. The tuple consists of the required child mastery threshold and the required robot level threshold. The required threshold for children's mastery directly points to a specific dimension in the child's growth profile, with each skill node corresponding to a dimension. The required robot level threshold is a positive integer, and its value corresponds to the level depth of the skill node in the skill tree.

7. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 1, characterized in that, The causal response layer records the embodied changes that the robot directly produces due to the changes in the child's growth profile, including the ability field group, the growth scale field group, and the narrative progress field group; The derived statistics layer uses the historical sequence of the robot's life form profile in the most recent 30 interactions as the data source to perform periodic batch derived calculations, including growth rate, growth stability, emotion recovery half-life, interaction type preference vector, and effective arousal period.

8. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 7, characterized in that, The ability field group includes a skill node status table, a learned action library index table, a learned voice library index table, and a learned lighting effect mode index table; the skill node status table stores the current status of each of all preset skill nodes, which is either locked or unlocked. The growth scale field group includes two dimensions: level and growth stage. The positive integer value of the level is kept in sync with the level field in the robot life form profile. This field is updated synchronously each time the level in the profile is improved. The narrative progress field group contains two dimensions: story progress and world view nodes. The story progress is an integer variable that records the sequence number of the completed story nodes in the current story sequence, with an initial value of 0. The world view nodes are string identifiers that point to the projection scene resource that should be displayed at the moment.

9. The AI ​​personality generation method for adaptive companionship interaction in children according to claim 1, characterized in that, In step S4, a decision-making mechanism is executed, which includes: interaction intention decision, interaction strategy selection, and emotion expression intensity decision. The interaction intent decision-making process combines focus score, fatigue score, and emotional polarity score to form a comprehensive demand index; the comprehensive demand index is calculated as follows: ;in, It is a comprehensive demand index; It is a focus score; It is a fatigue score; It is an emotion polarity mapping score, which is obtained by linear mapping from the emotion polarity score; at the same time, it reads the effective wake-up period in the statistical layer derived from the robot life form state model to determine whether the current moment is within the effective wake-up period; If the overall demand index is greater than 55, the current interaction has been more than 90 seconds since the last interaction, and the current time is within the effective wake-up period or the current time period has been marked as allowed by the parental rules, then the arbitrator outputs the intention to initiate an interaction; if the fatigue score is greater than 70 and the current interaction duration has exceeded 15 minutes, then the arbitrator outputs the intention to suggest rest; if the overall demand index is between 30 and 55, then the arbitrator outputs the intention to provide silent companionship; if the overall demand index is less than 30, then the arbitrator outputs the intention not to disturb. The interaction strategy selection is executed only when the interaction is initiated actively; based on the long-term percentile ranking of each dimension in the child's growth profile, the three dimensions with the lowest current percentile are identified as weak dimensions; at the same time, the interaction type preference vector in the derived statistical layer of the robot life form state model is read, and the two interaction types with the highest frequency are identified as preference types. If the initiative score is below 40, the strategy is encouragement-based; if the initiative score is between 40 and 70, the strategy is guidance-based; if the initiative score is above 70, the strategy is exploration-based. The emotion expression intensity decision is executed under all intention types; based on the emotion polarity score and the current mood value in the robot's life form profile, the dominant emotion type and expression intensity of the robot's current interaction are jointly determined.

10. An AI personality generation system for adaptive companionship interaction with children, applied to the AI ​​personality generation method for adaptive companionship interaction with children as described in any one of claims 1 to 9, characterized in that, The system includes: The perception and recognition module is used to synchronously collect multi-dimensional raw signals from children within a preset sampling period to form an interaction record vector. Based on the interaction record vector, time window aggregation and feature calculation are performed to obtain the child's instantaneous state vector and the current interaction event type label. The multi-dimensional raw data includes a sequence of frontal facial images, three-dimensional spatial coordinates, voice signals, touch event data, relative motion data, and timestamps. The child's instantaneous state vector is a sliding window formed by taking all interaction record vectors within the previous 30 seconds as the endpoint, and calculating the child's state indicators. The state indicators include focus score, fatigue score, emotional polarity score, interaction initiative score, and physical activity intensity score. The profile update module receives the child's real-time state vector and interaction event type label from the perception and recognition module. Based on the child's real-time state vector and interaction event type label, and taking the current interaction event as the basis, it performs synchronous incremental updates and outputs the updated child growth profile and robot life form profile. The coupling mapping module is used to receive the updated child growth profile and robot life form profile output by the profile update module; read all preset skill nodes in the robot skill tree; perform bidirectional coupling mapping operation based on the updated child growth profile and robot life form profile; output skill node unlocking instruction set; update the robot life form state model using the unlocking instruction set; the robot life form state model includes a causal response layer and a derived statistics layer. The personality arbitration module receives the child's real-time state vector output by the perception and recognition module, the updated robot life form portrait output by the portrait update module, the derived statistical layer fields in the updated robot life form state model output by the coupling mapping module, and a set of configurable personality parameters; it executes the decision-making mechanism to obtain the interaction intention, interaction strategy, and intensity of emotion expression. The interaction generation module receives the skill node unlocking instruction set output by the coupling mapping module and the interaction intent, strategy type, and emotional expression intensity output by the personality arbitration module; it then executes three sub-tasks in sequence: skill unlocking execution, achievement judgment, and interaction content synthesis to generate the final interaction content and instruction operation.

Images