A home intelligent old-age service robot system

Abstract

The application relates to a home intelligent old-age service robot system, belonging to the field of artificial intelligence and robot technology, which comprises a multi-modal perception and characteristic unit, which is used for collecting multi-modal original data of a user and an environment and processing the multi-modal original data into multi-modal characteristic vectors; a latent demand intensity evaluation unit, which is used for receiving the multi-modal characteristic vectors and calculating a latent demand comprehensive intensity; meanwhile, the unit integrates the received multi-modal characteristic vectors and outputs a fusion characteristic vector for use by a subsequent unit; the latent demand intensity evaluation unit also compares the latent demand comprehensive intensity with a preset demand triggering threshold value, generates a demand triggering signal when the latent demand comprehensive intensity is greater than the demand triggering threshold value, and maintains a continuous monitoring state when the latent demand comprehensive intensity is not greater than the demand triggering threshold value; and the application greatly improves the timeliness of service and the prevention ability of potential risks.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and robotics technology, and specifically to a home intelligent elderly care service robot system. Background Art

[0002] Existing home service robots primarily rely on explicit user commands to operate. They lack the ability to effectively recognize and respond to the implicit needs of elderly users, who for various reasons are unable or unwilling to clearly express them, such as subtle posture changes caused by discomfort or unconscious sounds of thirst. This lack of capability leads to delays, missed calls, and even unnecessary interruptions due to misjudgments, impacting their practical value and user acceptance. This invention aims to address this technical pain point by proposing an intelligent system that can proactively, accurately, and appropriately respond to users' implicit needs.

[0003] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0004] The purpose of the present invention is to provide a home intelligent elderly care service robot system to solve the problems raised in the above background technology.

[0005] The technical solution of the present invention is to include: a multimodal sensing and characterization unit for collecting multimodal raw data of the user and the environment, and processing the multimodal raw data into a multimodal feature vector;

[0006] The potential demand strength assessment unit is used to receive multimodal feature vectors to calculate the comprehensive strength of potential demand. At the same time, the unit integrates the received feature vectors of each modality and outputs a fused feature vector for use by subsequent units. The potential demand strength assessment unit also compares the comprehensive strength of potential demand with a preset demand trigger threshold. When the comprehensive strength of potential demand is greater than the demand trigger threshold, a demand trigger signal is generated. When the comprehensive strength of potential demand is not greater than the demand trigger threshold, the system maintains a continuous monitoring state.

[0007] The service appropriateness decision unit is used to respond to the demand trigger signal and calculate the comprehensive appropriateness score of the candidate service sequence based on the fused feature vector generated by the potential demand intensity assessment unit and the current user status;

[0008] The service selection and execution unit is used to select and execute the candidate service sequence with the highest comprehensive appropriateness score;

[0009] The parameter adaptive optimization unit is used to optimize the demand trigger threshold online based on user feedback on service execution.

[0010] Preferably, the process of calculating the comprehensive strength of potential demand by the potential demand strength evaluation unit is as follows:

[0011] Based on the multimodal feature vector, the abnormality score corresponding to each mode is calculated; the abnormality score of each mode is weighted and summed to obtain the initial intensity value; the situational adjustment factor is calculated according to the environmental data in the multimodal original data; the initial intensity value and the situational adjustment factor are combined to perform nonlinear fusion processing to generate the comprehensive intensity of potential demand.

[0012] Preferably, the online optimization process of the parameter adaptive optimization unit for the demand trigger threshold is as follows:

[0013] When receiving user rejection feedback for proactive services, increase the demand trigger threshold;

[0014] When system analysis identifies missed service opportunities, lower the demand trigger threshold.

[0015] Preferably, the process of calculating the comprehensive suitability score by the service suitability decision unit is as follows:

[0016] Based on the fused feature vector, the intention matching degree of the user in each predefined potential intention is calculated; based on the intention matching degree and the preset service utility matrix, the expected service utility is calculated; based on the current user status, the user interference cost is calculated through the rule set; and a comprehensive appropriateness score is generated by combining the expected service utility and the user interference cost.

[0017] Preferably, the service utility matrix is ​​used to store the efficiency of each candidate service sequence in solving each predefined potential intention; the parameter adaptive optimization unit is also used to perform online learning and update the service utility matrix based on user feedback.

[0018] Preferably, the user interference cost is dynamically calculated by a rule set, and the rule set is used to determine the interruption cost of executing the service according to the user's activity scenario.

[0019] Preferably, the rule set includes:

[0020] When the multimodal perception and characterization unit detects that the user is in a social scene or a focused scene, a high user interference cost is set;

[0021] When the multimodal perception and characterization unit detects that the user is in an idle scenario, a low user interference cost is set.

[0022] Preferably, the rule set further includes an emergency amendment clause, which is used to:

[0023] When the comprehensive intensity of potential demand exceeds the preset emergency threshold, the user interference cost is forcibly set to zero.

[0024] The present invention provides a home intelligent elderly care service robot system through improvement, which has the following improvements and advantages compared with the existing technology:

[0025] 1. The present invention incorporates a potential demand intensity assessment unit, not present in existing technologies. This unit proactively and continuously integrates scattered data from the multimodal sensing and characterization unit into a quantitative indicator that can directly trigger service decisions. This enables the system to understand users' implicit needs, transcending the passive model that relies on explicit instructions, achieving predictive and proactive service delivery, and effectively addressing the issue of insufficient demand expression among elderly users. The system uses the multimodal sensing and characterization unit to capture visual signals such as users' unconscious body curling and slight tremors. Combined with the drop in room temperature detected by the environmental sensor, the potential demand intensity assessment unit calculates a high comprehensive potential demand intensity. When this value exceeds the dynamically adjusted demand trigger threshold, the system proactively initiates service. This proactive approach significantly improves the timeliness of service and the ability to prevent potential risks.

[0026] 2. This invention uniquely creates a decision-making framework based on a service appropriateness decision-making unit, specifically introducing a new dimension: user interference cost. While existing technologies can predict user needs through certain rules, their service behaviors are often rigid and context-insensitive. This invention dynamically assesses the interference cost of performing services in different scenarios, such as when users are socializing, reading, or relaxing, thereby enabling opportune service execution. This careful consideration of interaction timing is a major design flaw in existing technologies.

[0027] 3. This invention uses a parameter adaptive optimization unit to give the entire system the ability to learn and evolve. While the service logic in existing technologies is typically fixed and unchanging, the system of this invention can optimize demand trigger thresholds and service utility matrices online based on user rejection or acceptance feedback. This enables the robot to gradually adapt to the unique habits and preferences of a single user, achieving a profound shift from general services to personalized services. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The present invention will be further explained below in conjunction with the accompanying drawings and examples:

[0029] Figure 1 It is a flow chart of the system of the present invention. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0031] Example 1:

[0032] See also Figure 1 ,The present invention provides a home intelligent elderly care service robot system, comprising: a multimodal perception and characterization unit, for collecting multimodal raw data of a user and an environment, and processing the multimodal raw data into a multimodal feature vector;

[0033] The potential demand strength assessment unit is used to receive multimodal feature vectors to calculate the comprehensive strength of potential demand. At the same time, the unit integrates the received feature vectors of each modality and outputs a fused feature vector for use by subsequent units. The potential demand strength assessment unit also compares the comprehensive strength of potential demand with a preset demand trigger threshold. When the comprehensive strength of potential demand is greater than the demand trigger threshold, a demand trigger signal is generated. When the comprehensive strength of potential demand is not greater than the demand trigger threshold, the system maintains a continuous monitoring state.

[0034] The service appropriateness decision unit is used to respond to the demand trigger signal and calculate the comprehensive appropriateness score of the candidate service sequence based on the fused feature vector generated by the potential demand intensity assessment unit and the current user status;

[0035] The service selection and execution unit is used to select and execute the candidate service sequence with the highest comprehensive appropriateness score;

[0036] The parameter adaptive optimization unit is used to optimize the demand trigger threshold online based on user feedback on service execution.

[0037] A home-based intelligent elderly care service robot system establishes a closed technical loop from perception, assessment, decision-making, to execution and optimization. The system integrates a multimodal perception and characterization unit, a potential demand intensity assessment unit, a service appropriateness decision unit, a service selection and execution unit, and a parameter adaptive optimization unit. This system aims to address the technical issue of existing service robots' inability to respond to the implicit needs of elderly users. The system's workflow begins with the multimodal perception and characterization unit continuously collecting and characterizing user and environmental status data. The potential demand intensity assessment unit quantitatively analyzes the acquired data and compares the results with a demand trigger threshold dynamically adjusted by the parameter adaptive optimization unit. This process forms the basis for determining service initiation. Once a demand is confirmed, the service appropriateness decision unit immediately intervenes, comprehensively evaluating the service content and execution timing. The service selection and execution unit then completes the designated service action. The endpoint of this process also marks the starting point for a new round of optimization, forming an adaptive system that can continuously improve itself based on user feedback to provide more predictable, personalized, and appropriate elderly care services, significantly improving the effectiveness of the robot in real home environments.

[0038] Example 2:

[0039] The process of calculating the comprehensive strength of potential demand by the potential demand strength assessment unit is as follows:

[0040] Based on the multimodal feature vector, the abnormality score corresponding to each mode is calculated; the abnormality scores of each mode are weighted and summed to obtain the initial strength value; the context adjustment factor is calculated based on the environmental data in the multimodal raw data; the initial strength value and the context adjustment factor are combined and nonlinearly fused to generate the comprehensive strength of potential demand;

[0041] The online optimization process of the parameter adaptive optimization unit for the demand trigger threshold is as follows:

[0042] When receiving user rejection feedback for proactive services, increase the demand trigger threshold;

[0043] When system analysis identifies missed service opportunities, lower the demand trigger threshold.

[0044] To integrate discrete, multi-source physiological and behavioral cues from users with environmental factors into a unified quantitative indicator, the potential demand intensity assessment unit uses a heuristic fusion model. The technical motivation for this model is that the user's potential demand intensity is not only a weighted synthesis of their own multiple abnormal signals, but also that this comprehensive intensity should be dynamically affected by the comfort level of the environment in which they are located. The mathematical model is constructed as follows:

[0045] ;

[0046] in, Indicates the comprehensive strength of potential demand, which is a dimensionless floating point value; Represents a collection of perceptual modalities, such as gesture, expression, and voice; is any perceptual modality in the set and serves as the sum index; Indicates modality The corresponding dimensionless weights are initially set by domain experts and can be optimized online later. ;

[0047] Its online optimization process can adopt a feedback-based heuristic adjustment rule: when the system triggers a service and the user accepts the service, or the system confirms that the service is effective through analysis, the system will give a higher abnormality score to the user. The mode of the corresponding weight will be moderately increased, and conversely, when the service is explicitly rejected by the user, the modal weights that contribute to higher abnormality scores will be moderately reduced; for example, the update rule can be ;or ;in is a small learning rate, and after each update, all weights must be renormalized to ensure that their sum is always 1;

[0048] From the modal The original feature vector extracted from ; : Context sensitivity coefficient, a dimensionless parameter used to adjust the degree of influence of the context adjustment factor. Its value range is between [0, 1] and can be adjusted according to the importance of environmental factors on users in actual application scenarios; : situational adjustment factor, a dimensionless value in the interval [-1, 1], used to quantify the impact of environmental comfort;

[0049] A feasible feature extraction method is: for posture modality, the image coordinates of 18 key skeletal joints of the human body, such as head, neck, shoulder, elbow, wrist, hip, knee, ankle, etc., can be obtained through the onboard 2D camera, and all coordinate points are normalized relative to the length of the line connecting the center point of the neck and hip to form a 36-dimensional feature vector For facial expression modality, facial key point detection technology can be used to calculate the intensity values ​​of facial behavior coding units, such as AU1 (inner eyebrow raising), AU4 (eyebrow lowering), AU12 (mouth corner raising), etc., to form a feature vector containing several key AU intensity values For the sound modality, the collected user sound signal can be segmented into 25 milliseconds per frame and 10 milliseconds per frame shift, and the 13-dimensional Mel frequency cepstral coefficients of each frame can be extracted to form a feature vector ;

[0050] It is the combination of different physical dimensions A normalized function that maps the dimensionless anomaly score to a uniform value in the interval [0, 1];

[0051] The normalization function One way to achieve this is to use anomaly detection based on Mahalanobis distance. First, at the initial stage of system deployment, multiple sets of feature vectors of users in a comfortable and no-clear-demand state are collected to form a normal state data set for each mode; based on this data set, the normal state of each mode is calculated. The characteristic mean vector of and covariance matrix ; For the newly collected feature vector ; The square of Mahalanobis distance is:

[0052] ;

[0053] in, : eigenvector The Mahalanobis distance, : represents the transpose of a vector; : represents the inverse of the matrix, and the anomaly score maps the distance to the [0, 1] interval through a logical function, for example:

[0054] ;

[0055] in, and It is the preset scaling and translation parameters used to adjust the steepness and center point position of the function curve. Its value can be determined based on empirical data to obtain the best differentiation effect; : Anomaly score, which is a value that maps the Mahalanobis distance to the interval [0, 1] through the logistic function; : the base of natural logarithms; : Mahalanobis distance;

[0056] It represents the situational adjustment factor, which is a dimensionless value in the interval [-1, 1] calculated based on the physical parameters of the environment and is used to quantify the environmental impact;

[0057] The ambient temperature , degrees Celsius and ambient light intensity , Lux and other key physical parameters are weighted and calculated by hyperbolic tangent function Normalize; for example, set the comfort temperature reference value to The reference value for suitable light is , the calculation formula can be:

[0058] ;

[0059] in, and is the sensitivity weight of different environmental parameters, : Temperature sensitivity weight : Light sensitivity weight, for example, you can set ℃ and ; This formula makes the environment worse when the ambient temperature or light intensity is lower than the reference value. If it is positive, it will amplify the potential demand intensity; otherwise, it will be negative, which will have a suppressive effect; : Reference value of comfortable temperature, in °C; : actual ambient temperature, in °C; : Reference value of suitable light, unit is ; : The actual ambient light intensity, in units of ;

[0060] is the context sensitivity coefficient, a dimensionless parameter used to adjust the extent of the impact;

[0061] In the application, the potential demand intensity assessment unit can accurately quantify the user's non-explicit needs through this formula; for example, when the user's curled-up posture is detected, The value is high and the indoor temperature is low. When it is a positive value, the calculated will be significantly amplified, exceeding the demand trigger threshold ; The initial value of is determined based on the ROC curve analysis of the labeled data set to balance the sensitivity and specificity of the service; the parameter adaptive optimization unit makes online adjustments based on user feedback on the service: if the service is rejected, the value is increased. To reduce the sensitivity of the system; if the subsequent analysis finds that there are service omissions, then reduce To improve sensitivity; this mechanism ensures that the system can dynamically optimize and achieve precise control of the timing of service initiation.

[0062] Example 3:

[0063] The process of calculating the comprehensive appropriateness score by the service appropriateness decision unit is as follows:

[0064] Based on the fused feature vector, the user's intention matching degree for each predefined potential intention is calculated. Based on the intention matching degree and the preset service utility matrix, the expected service utility is calculated. Based on the current user status, the user interference cost is calculated using a set of rules. The expected service utility and user interference cost are combined to generate a comprehensive appropriateness score.

[0065] The service utility matrix is ​​used to store the efficiency of each candidate service sequence in solving each predefined potential intention; the parameter adaptive optimization unit is also used to perform online learning and update the service utility matrix based on user feedback.

[0066] After the potential demand strength assessment unit confirms the existence of demand, the service appropriateness decision unit is activated to determine the specific service content and execution timing. This decision-making process is based on an evaluation model improved to enhance the human-computer interaction experience. The design concept of this model is derived from the expected utility theory of decision theory and uniquely introduces the interference cost of service execution timing to users as a key decision factor, thereby ensuring that the decision takes into account both content effectiveness and timing appropriateness. The mathematical model is constructed as follows:

[0067] ;

[0068] in, Represents a candidate service sequence At the time point The comprehensive appropriateness score is the dimensionless output of the model; is the candidate service sequence identifier; is a candidate execution time point; is the index used to traverse the collection, It is a set of predefined potential intents; is a collection any specific intention in The fusion feature vector is composed of the original feature vectors of each modality received by the potential demand strength assessment unit. It integrates all the underlying original information that triggers the service demand and provides a comprehensive input for subsequent intent inference. Its composition reflects the cascade logical relationship between models. : fused feature vector, which is formed by concatenating the original feature vectors of each modality;

[0069] The fused feature vector is the original eigenvector of each mode used by the potential demand strength assessment unit The system is composed of three modes: gesture, expression and voice.

[0070] ;

[0071] in, : fusion feature vector; : Feature vector extracted from the posture modality; : Feature vector extracted from the facial expression modality; : A feature vector extracted from the sound modality; this vector integrates all the underlying raw information that triggers the service demand, providing comprehensive input for subsequent intent inference;

[0072] Indicates the intent matching degree, which is the posterior probability output by the pre-trained classifier;

[0073] The pre-trained classifier can be a multi-layer perceptron neural network, the number of neurons in the input layer of the network is equal to the fusion feature vector The number of neurons in the output layer is equal to the predefined potential intent set The total number of intentions in the output layer; the output layer uses the Softmax activation function to ensure that its output value is a legal probability distribution, representing the posterior probability of the user having various potential intentions under the current observation; the classifier is trained through supervised learning with a pre-labeled dataset, where each Each sample corresponds to a real user intent label ;

[0074] Represents service utility, which is a dimensionless value in the interval [0, 1] stored in the service utility matrix. Intention Solution efficiency; is the user interference cost, which is a dimensionless value in the interval [0, 1];

[0075] In the application, the service appropriateness decision unit is based on vector, calculate the probability distribution of the user's potential intention; query the service utility matrix to identify the high probability intention The parameter adaptive optimization unit will adjust the service utility matrix according to the user feedback after the service is executed. The value is updated through online learning. If the service successfully solves the user's intention, the corresponding utility value is increased.

[0076] The online learning update can adopt a Q-learning update mechanism similar to reinforcement learning, assuming that the learning rate is , when the service sequence Executed in response to inferred intent Afterwards, if the system receives positive user feedback or confirms the success of the task through sensors, it can be regarded as a reward of Reward = 1, and the utility value is updated to:

[0077] ;

[0078] If negative feedback is obtained (reward = 0), the update is:

[0079] ;

[0080] in, : updated service utility value; : old service utility value before update; : learning rate, a value in the interval (0, 1); : The service sequence to be executed; : The user intention inferred by the system; this mechanism makes the utility value of successful services approach 1, while the utility value of failed services approaches 0, thus achieving adaptive optimization of the utility matrix

[0081] This mechanism ensures the effectiveness of service recommendations and improves personalization and accuracy through continuous learning.

[0082] Example 4:

[0083] The user interference cost is dynamically calculated by a rule set, which is used to determine the interruption cost of executing the service based on the user's activity scenario;

[0084] The rule set includes:

[0085] When the multimodal perception and characterization unit detects that the user is in a social scene or a focused scene, a high user interference cost is set;

[0086] When the multimodal perception and characterization unit detects that the user is in an idle scenario, a low user interference cost is set;

[0087] The rules also include emergency amendments, which are used to:

[0088] When the comprehensive intensity of potential demand exceeds the preset emergency threshold, the user interference cost is forcibly set to zero.

[0089] The calculation of user interference cost is a key step in achieving human-centered interaction design in this system. Its value is dynamically determined by a set of clear and enforceable rules based on the user's real-time activity. This set of rules is designed to carefully assess the potential disruption of service interventions and ensure natural interaction.

[0090] To further clarify, the rule set is determined based on the detection results of the multimodal perception and characterization units; when the speech recognition module detects that the user is in a continuous conversation, the system determines that the user is in a social scene and sets a high user interference cost. , for example, 0.95; when the visual module recognizes that the user's gaze is focused on a specific object, such as a book, for a long time and the posture is stable, the system determines that the user is in a focused scene and sets a higher user interference cost, such as 0.7; in contrast, when no specific activity scene is matched, the system determines that the user is in an idle scene, and the user interference cost is set to a low value, such as 0.1; through this scenario-based differentiated cost setting, the system can make refined selections of service opportunities;

[0091] To ensure safe response in extreme situations, the rule set has an emergency correction clause built in. The setting logic of this clause is that when the output of the potential demand intensity assessment unit Exceeding a preset emergency threshold This clause is activated when the user is interrupted; the value of the emergency threshold is determined based on statistical analysis of data on known emergency events, such as falls, to ensure that it can reliably indicate high-risk conditions; once activated, this clause will force the user to interrupt the cost Set to zero; this makes the timing factor in the appropriateness evaluation formula The value of is always 1, thus ensuring that the execution priority of emergency rescue services is not affected by any interference factors; the existence of this amendment clause ensures that the system takes into account the necessary safety redundancy while pursuing interactive comfort.

[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A home intelligent elderly care service robot system, characterized in that: include: Multimodal perception and characterization unit, used to collect multimodal raw data of users and the environment, and process the multimodal raw data into multimodal feature vectors; The potential demand strength evaluation unit is used to receive the multimodal feature vector to calculate the comprehensive strength of potential demand. The calculation formula is as follows: ； in, Indicates the comprehensive strength of potential demand, which is a dimensionless floating point value; represents a collection of perceptual modalities; is any perceptual modality in the set and serves as the sum index; Indicates modality The corresponding dimensionless weight, and ; From the modal The original feature vector extracted from ; : Situational sensitivity coefficient, a dimensionless parameter used to adjust the degree of influence of the situational adjustment factor, with a value range of [0, 1]. : situational adjustment factor, a dimensionless value in the interval [-1, 1], used to quantify the impact of environmental comfort; It is the combination of different physical dimensions A normalized function that maps the dimensionless anomaly score to a uniform value in the interval [0, 1]; At the same time, the unit integrates the received feature vectors of each modality and outputs a fused feature vector for use by subsequent units. The potential demand strength assessment unit also compares the comprehensive strength of potential demand with the preset demand trigger threshold. When the comprehensive strength of potential demand is greater than the demand trigger threshold, a demand trigger signal is generated. When the comprehensive strength of potential demand is not greater than the demand trigger threshold, the system maintains a continuous monitoring state. The service appropriateness decision unit is used to respond to the demand trigger signal and calculate the comprehensive appropriateness score of the candidate service sequence based on the fused feature vector generated by the potential demand intensity assessment unit and the current user status. The calculation formula is as follows: ； in, Represents a candidate service sequence At the time point The comprehensive appropriateness score is the dimensionless output of the model; is the candidate service sequence identifier; is a candidate execution time point; is the index used to traverse the collection, It is a set of predefined potential intents; is a collection any specific intention in To fuse the feature vector, the vector is formed by splicing the original feature vectors of each mode received by the potential demand strength assessment unit. Indicates the degree of intention matching, which is the posterior probability output by the pre-trained classifier; the pre-trained classifier can be a multi-layer perceptron neural network, the number of neurons in the input layer of the network is the same as the fusion feature vector The number of neurons in the output layer is equal to the predefined potential intent set The total number of intents in ; Represents service utility, which is a dimensionless value in the interval [0, 1] stored in the service utility matrix. Intention Solution efficiency; is the user interference cost, which is a dimensionless value in the interval [0, 1]; The service selection and execution unit is used to select and execute the candidate service sequence with the highest comprehensive appropriateness score; The parameter adaptive optimization unit is used to optimize the demand trigger threshold online based on user feedback on service execution.

2. The intelligent home elderly care service robot system according to claim 1, characterized in that: The process of calculating the comprehensive strength of potential demand by the potential demand strength assessment unit is as follows: Based on the multimodal feature vector, the abnormality score corresponding to each mode is calculated; the abnormality score of each mode is weighted and summed to obtain the initial intensity value; Calculate contextual adjustment factors based on environmental data in multimodal raw data; The initial intensity value and the situational adjustment factor are combined to perform nonlinear fusion processing to generate the comprehensive intensity of potential demand.

3. The intelligent home elderly care service robot system according to claim 1, characterized in that: The online optimization process of the parameter adaptive optimization unit for the demand trigger threshold is as follows: When receiving user rejection feedback for proactive services, increase the demand trigger threshold; When system analysis identifies missed service opportunities, lower the demand trigger threshold.

4. The intelligent home elderly care service robot system according to claim 1, characterized in that: The process of calculating the comprehensive appropriateness score by the service appropriateness decision unit is as follows: Based on the fused feature vector, the intention matching degree of the user in each predefined potential intention is calculated; based on the intention matching degree and the preset service utility matrix, the expected service utility is calculated; based on the current user status, the user interference cost is calculated through the rule set; and a comprehensive appropriateness score is generated by combining the expected service utility and the user interference cost.

5. The intelligent home elderly care service robot system according to claim 4, characterized in that: The service utility matrix is ​​used to store the efficiency of each candidate service sequence in solving each predefined potential intention; the parameter adaptive optimization unit is also used to perform online learning and update the service utility matrix based on user feedback.

6. The intelligent home elderly care service robot system according to claim 4, characterized in that: The user interference cost is dynamically calculated by a rule set, and the rule set is used to determine the interruption cost of executing a service based on the user's activity scenario.

7. The intelligent home elderly care service robot system according to claim 6, characterized in that: The rule set includes: When the multimodal perception and characterization unit detects that the user is in a social scene or a focused scene, a high user interference cost is set; When the multimodal perception and characterization unit detects that the user is in an idle scenario, a low user interference cost is set.

8. The intelligent home elderly care service robot system according to claim 7, characterized in that: The rules also include emergency amendments, which are used to: When the comprehensive intensity of potential demand exceeds the preset emergency threshold, the user interference cost is forcibly set to zero.

Images