Interaction method and system of pension service robot and storage medium

Abstract

This invention relates to the field of elderly care service technology, and discloses an interaction method, system, and storage medium for an elderly care service robot. The method includes the following steps: S1, in response to a triggered interaction event, acquiring the current state data of the target elderly person. The current state data includes physiological data and environmental data collected by the robot's sensors, as well as a demand description text collected in response to a voice wake-up command in the interaction event; S2, based on a preset elderly care service knowledge base, performing intent recognition on the current state data to determine the current service demand type of the target elderly person. The preset elderly care service knowledge base contains multiple service demand types and a state feature set corresponding to each service demand type. This invention, by responding to interaction events and integrating the collected physiological data, environmental data, and demand description text of the elderly person, achieves multimodal data fusion, providing a rich data foundation for accurately identifying service demands and significantly improving the robot's ability to perceive the elderly person's state.

Description

Technical Field

[0001] This invention relates to the field of elderly care service technology, specifically to the interaction method, system, and storage medium of elderly care service robots. Background Technology

[0002] With the accelerating aging of the population, the demand for elderly care services is increasing. Elderly care service robots have been widely used as an important means to solve the shortage of elderly care resources. However, most existing elderly care service robots adopt a fixed response mode, relying solely on voice commands for interaction. This results in problems such as low recognition accuracy and rigid response content. For example, when an elderly person feels dizzy due to high blood pressure and describes "chest tightness," the robot can only recognize the keyword "chest tightness" and cannot combine it with physiological data to judge the underlying health risks. In addition, the rigid response content of existing robots cannot be dynamically adjusted according to the elderly person's personalized characteristics and real-time feedback, resulting in a monotonous interactive experience that fails to meet the growing personalized and emotional needs of the elderly. Therefore, how to improve the ability of elderly care service robots to accurately identify the needs of the elderly and their personalized interaction capabilities has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0003] The purpose of this invention is to provide an interaction method, system, and storage medium for elderly care service robots to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an interaction method for elderly care service robots, comprising the following steps:

[0005] S1. In response to the triggered interaction event, obtain the current status data of the target elderly person. The current status data includes physiological data and environmental data collected by the robot's sensors, as well as the demand description text collected in response to the voice wake-up command in the interaction event.

[0006] S2. Based on the preset elderly care service knowledge base, perform intent recognition on the current status data to determine the current service demand type of the target elderly. The preset elderly care service knowledge base contains multiple service demand types and the status feature set corresponding to each service demand type.

[0007] S3. Based on the current service demand type, match the corresponding initial response template from the preset multimodal response strategy library, and use the current status data to populate the content of the initial response template to generate personalized response content.

[0008] S4. Control the robot to execute personalized response content and simultaneously start one or more auxiliary service devices associated with the current service request type;

[0009] S5. Continuously monitor the feedback data of the target elderly during the implementation of personalized response content and assistive services;

[0010] S6. Update the preset elderly care service knowledge base based on the feedback data. The updated preset elderly care service knowledge base will be called in subsequent intent recognition steps.

[0011] As a preferred technical solution of the present invention, the interactive event responsive to the trigger includes: responsive to the trigger event, wherein the trigger event is selected from at least one of the following: a voice wake-up event initiated by the target elderly person, an abnormal physiological sign event detected by the robot's sensors, and an environmental safety alarm event detected by the robot's sensors.

[0012] As a preferred embodiment of the present invention, the step of performing intent recognition on current state data based on a preset elderly care service knowledge base to determine the current service demand type of the target elderly person includes:

[0013] Match the current status data with the status feature sets corresponding to each service demand type in the preset elderly care service knowledge base;

[0014] The state feature set is a multi-dimensional feature vector, which includes at least physiological indicator threshold ranges, environmental parameter ranges, and keyword weight tables; the matching uses a weighted Euclidean distance algorithm to calculate the matching degree between the current state data and each state feature set.

[0015] When the matching degree exceeds the preset threshold, the service requirement type with the highest matching degree will be determined as the current service requirement type;

[0016] When the matching degree does not exceed the preset threshold, the current status data is marked as data to be confirmed, and a confirmation request is sent to the preset monitoring terminal.

[0017] As a preferred embodiment of the present invention, the step of matching a corresponding initial response template from a preset multimodal response strategy library according to the current service demand type includes:

[0018] The initial response template is a formatted string containing named placeholders;

[0019] When the current service request type is emotional companionship, match an initial response template that includes requests for story playback, music playback, or video call.

[0020] When the current service request type is life assistance, match an initial response template that includes task breakdown guidance and progress inquiries;

[0021] When the current service request type is a security alert, match the initial response template that includes reassurance prompts and emergency contact connection requests.

[0022] As a preferred embodiment of the present invention, the synchronous activation of one or more auxiliary service devices associated with the current service demand type includes:

[0023] The synchronous startup is achieved by sending control commands to the auxiliary service device through a preset communication protocol. The control commands include device identifier, operation type, and operation parameters.

[0024] When the current service request type is health monitoring, the data upload channel of the remote medical terminal is activated simultaneously.

[0025] When the current service request type is "life assistance", the operation permissions of smart home devices will be activated simultaneously.

[0026] When the current service request type is a security alert, a verification command for the indoor lighting system and the lock status of doors and windows is simultaneously initiated.

[0027] As a preferred embodiment of the present invention, the continuous monitoring of feedback data from the target elderly person during the execution of personalized response content and assistive services includes:

[0028] The robot collects facial expression data and voice emotion data of the target elderly person through its sensors.

[0029] Obtain service execution status data through auxiliary service devices;

[0030] A pre-defined correlation analysis model is used to perform correlation analysis on the facial expression data, voice emotion data, and service execution status data to generate a quantitative comprehensive feedback evaluation result; wherein, the correlation analysis model is a logistic regression model, and the output comprehensive feedback evaluation result is a satisfaction score between 0 and 1.

[0031] As a preferred embodiment of the present invention, the step of updating the preset elderly care service knowledge base based on feedback data includes:

[0032] Based on the comprehensive feedback evaluation results, an update strategy is determined: if the satisfaction score exceeds the preset update threshold, the current state data of this interaction and the corresponding service demand type are used as positive samples and added to the state feature set of the corresponding service demand type in the preset elderly care service knowledge base to optimize subsequent intent recognition.

[0033] The interactive system of the elderly care service robot includes:

[0034] The data acquisition module is configured to acquire the current state data of the target elderly person in response to a triggered interactive event. The current state data includes physiological data and environmental data collected by the robot's sensors, as well as a demand description text collected in response to a voice wake-up command in the interactive event.

[0035] The intent recognition module is configured to perform intent recognition on the current state data based on a preset elderly care service knowledge base to determine the current service demand type of the target elderly person. The preset elderly care service knowledge base contains multiple service demand types and a set of state features corresponding to each service demand type.

[0036] The response generation module is configured to match the corresponding initial response template from a preset multimodal response strategy library according to the current service demand type, and use the current status data to populate the content of the initial response template to generate personalized response content.

[0037] The service execution module is configured to control the robot to execute the personalized response content and simultaneously launch one or more auxiliary service devices associated with the current service request type;

[0038] The feedback collection module is configured to continuously monitor the feedback data of the target elderly person during the execution of the personalized response content and assistance services;

[0039] The knowledge base update module is configured to update the preset elderly care service knowledge base based on the feedback data. The updated preset elderly care service knowledge base is used to be invoked in subsequent intent recognition steps.

[0040] As a preferred embodiment of the present invention, the data acquisition module is configured to respond to a triggering event, wherein the triggering event is selected from at least one of the following: a voice wake-up event initiated by the target elderly person, an abnormal physiological sign event detected by the robot's sensors, and an environmental safety alarm event detected by the robot's sensors.

[0041] In a preferred embodiment of the present invention, the intent recognition module is configured as follows:

[0042] The current status data is matched with the status feature sets corresponding to each service demand type in the preset elderly care service knowledge base;

[0043] The state feature set is a multi-dimensional feature vector, which includes at least physiological indicator threshold ranges, environmental parameter ranges, and keyword weight tables; the matching uses a weighted Euclidean distance algorithm to calculate the matching degree between the current state data and each state feature set.

[0044] When the matching degree exceeds the preset threshold, the service requirement type with the highest matching degree will be determined as the current service requirement type;

[0045] When the matching degree does not exceed the preset threshold, the current status data is marked as data to be confirmed, and a confirmation request is sent to the preset monitoring terminal.

[0046] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the interaction method of the elderly care service robot as described above.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. This invention integrates physiological data, environmental data, and demand description text collected from the elderly by responding to interactive events, thereby achieving multimodal data fusion. This provides a rich data foundation for accurately identifying service needs and significantly improves the robot's ability to perceive the elderly's condition.

[0049] 2. This invention uses a pre-set elderly care service knowledge base to identify the intent of the current state data, thereby determining the current service needs of the target elderly person. The pre-set elderly care service knowledge base contains multiple service need types and corresponding state feature sets for each service need type. This knowledge base-based intent recognition mechanism enables the robot to comprehensively match multi-dimensional state data with the pre-set feature sets, overcoming the limitation of existing technologies that can only identify explicit keywords, and significantly improving the accuracy of service need identification. Especially when the elderly person's expression is unclear or the symptoms are complex, it can accurately identify implicit needs such as health monitoring and safety warnings.

[0050] 3. This invention matches the corresponding initial response template from a preset multimodal response strategy library according to the current service demand type, and fills the initial response template with content using the current state data to generate personalized response content. This template plus filling response generation method not only ensures the standardization of response content, but also realizes personalized adaptation to the current state data, enabling the robot's response to accurately respond to the current scenario. For example, it can provide specific numerical suggestions when blood pressure is abnormal, which significantly improves the pertinence and effectiveness of the interaction.

[0051] 4. This invention achieves coordinated interaction between interactive responses and auxiliary services by controlling a robot to execute personalized responses and simultaneously activating one or more auxiliary service devices associated with the current service request type. When a health monitoring request is identified, the robot not only outputs a personalized response but also automatically activates the data upload channel of the remote medical terminal to transmit physiological data to the medical center in real time. When a safety warning request is identified, the robot simultaneously activates the indoor lighting system and the verification instructions for the locking status of doors and windows. This linkage mechanism significantly shortens the response time for emergency events and provides more timely safety protection for the elderly.

[0052] 5. This invention continuously monitors the feedback data of the target elderly person during the execution of personalized response content and auxiliary services, and updates the preset elderly care service knowledge base based on the feedback data. The updated preset elderly care service knowledge base is used to call in subsequent intent recognition steps. This feedback-based knowledge base update mechanism enables the robot to have self-learning and self-optimization capabilities, and can continuously optimize the recognition strategy and response method based on historical interaction effects. As the usage time increases, the service quality and personalization of the robot continue to improve, avoiding the problem of declining interactive experience caused by the solidification of response strategies in existing technologies. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the overall process of the interaction method of the elderly care service robot of the present invention;

[0054] Figure 2 This is a structural block diagram of the interactive system of the elderly care service robot of the present invention. Detailed Implementation

[0055] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1

[0057] The interaction method for elderly care service robots includes the following steps:

[0058] S1. In response to the triggered interaction event, obtain the current status data of the target elderly person. The current status data includes physiological data and environmental data collected by the robot's sensors, as well as the demand description text collected in response to the voice wake-up command in the interaction event.

[0059] S2. Based on the preset elderly care service knowledge base, perform intent recognition on the current status data to determine the current service demand type of the target elderly. The preset elderly care service knowledge base contains multiple service demand types and the status feature set corresponding to each service demand type.

[0060] S3. Based on the current service demand type, match the corresponding initial response template from the preset multimodal response strategy library, and use the current status data to populate the content of the initial response template to generate personalized response content.

[0061] S4. Control the robot to execute personalized response content and simultaneously start one or more auxiliary service devices associated with the current service request type;

[0062] S5. Continuously monitor the feedback data of the target elderly during the implementation of personalized response content and assistive services;

[0063] S6. Update the preset elderly care service knowledge base based on the feedback data. The updated preset elderly care service knowledge base will be called in subsequent intent recognition steps.

[0064] Furthermore, the response to the triggered interactive event includes: a response to a triggering event, wherein the triggering event is selected from at least one of the following: a voice wake-up event initiated by the target elderly person, an abnormal physiological sign event detected by the robot's sensors, or an environmental safety alarm event detected by the robot's sensors.

[0065] Furthermore, based on a pre-defined elderly care service knowledge base, intent recognition is performed on the current status data to determine the type of current service needs of the target elderly person, including:

[0066] Match the current status data with the status feature sets corresponding to each service demand type in the preset elderly care service knowledge base;

[0067] The state feature set is a multi-dimensional feature vector, which includes at least the threshold range of physiological indicators, the range of environmental parameters, and a keyword weight table. The matching uses a weighted Euclidean distance algorithm to calculate the matching degree between the current state data and each state feature set. For example, the weight of physiological data can be set to 0.5, the weight of environmental data to 0.3, and the TF-IDF score weight of the keywords in the demand description text to 0.2. Then the matching degree can be calculated as 1 / (1+weighted Euclidean distance). When the matching degree exceeds the preset threshold, the service demand type with the highest matching degree is determined as the current service demand type.

[0068] When the matching degree does not exceed the preset threshold, the current status data is marked as data to be confirmed, and a confirmation request is sent to the preset monitoring terminal.

[0069] Furthermore, based on the current service request type, the corresponding initial response template is matched from the preset multimodal response strategy library, including:

[0070] The initial response template is a formatted string containing named placeholders, such as: "{name}, you are currently {symptom}, and {action} is suggested".

[0071] When the current service request type is emotional companionship, match an initial response template that includes requests for story playback, music playback, or video call.

[0072] When the current service request type is life assistance, match an initial response template that includes task breakdown guidance and progress inquiries;

[0073] When the current service request type is a security alert, match the initial response template that includes reassurance prompts and emergency contact connection requests.

[0074] Furthermore, simultaneously activate one or more auxiliary service devices associated with the current service demand type, including:

[0075] Synchronous startup is achieved by sending JSON-formatted control commands to auxiliary service devices through a preset communication protocol (such as MQTT protocol). The control commands include device identifier, operation type, and operation parameters.

[0076] When the current service request type is health monitoring, the data upload channel of the remote medical terminal is activated simultaneously.

[0077] When the current service request type is "life assistance", the operation permissions of smart home devices will be activated simultaneously.

[0078] When the current service request type is a security alert, a verification command for the indoor lighting system and the lock status of doors and windows is simultaneously initiated.

[0079] Furthermore, continuous monitoring of feedback data from the target elderly during the implementation of personalized responses and assistive services includes:

[0080] The robot collects facial expression data and voice emotion data of the target elderly person through its sensors.

[0081] Obtain service execution status data through auxiliary service devices;

[0082] A pre-defined correlation analysis model is used to perform correlation analysis on facial expression data, voice emotion data, and service execution status data to generate a quantitative comprehensive feedback evaluation result. The correlation analysis model is a logistic regression model, and the output comprehensive feedback evaluation result is a satisfaction score between 0 and 1.

[0083] Furthermore, the pre-set elderly care service knowledge base is updated based on feedback data, including:

[0084] Based on the comprehensive feedback evaluation results, the update strategy is determined as follows: if the satisfaction score exceeds the preset update threshold (e.g., 0.9), the current state data of this interaction and the corresponding service demand type are used as positive samples and added to the state feature set of the corresponding service demand type in the preset elderly care service knowledge base. This is to optimize the covariance matrix of the feature vector of this type, enhance the generalization ability of the model, and be used for subsequent intent recognition.

[0085] The interactive system of the elderly care service robot includes:

[0086] The data acquisition module is configured to acquire the current status data of the target elderly person in response to the triggered interaction event. The current status data includes physiological data and environmental data collected by the robot's sensors, as well as the demand description text collected in response to the voice wake-up command in the interaction event.

[0087] The intent recognition module is configured to perform intent recognition on the current state data based on a preset elderly care service knowledge base to determine the current service demand type of the target elderly person. The preset elderly care service knowledge base contains multiple service demand types and the state feature set corresponding to each service demand type.

[0088] The response generation module is configured to match the corresponding initial response template from the preset multimodal response strategy library according to the current service demand type, and use the current status data to populate the content of the initial response template to generate personalized response content.

[0089] The service execution module is configured to control the robot to execute personalized response content and simultaneously launch one or more auxiliary service devices associated with the current service request type;

[0090] The feedback collection module is configured to continuously monitor the feedback data of the target elderly person during the execution of personalized response content and assistive services;

[0091] The knowledge base update module is configured to update the preset elderly care service knowledge base based on feedback data. The updated preset elderly care service knowledge base is used to be called in subsequent intent recognition steps.

[0092] Furthermore, the data acquisition module is configured to respond to a triggering event, which is selected from at least one of the following: a voice wake-up event initiated by the target elderly person, an abnormal physiological sign event detected by the robot's sensors, or an environmental safety alarm event detected by the robot's sensors.

[0093] Furthermore, the intent recognition module is configured as follows:

[0094] Match the current status data with the status feature sets corresponding to each service demand type in the preset elderly care service knowledge base;

[0095] The state feature set is a multi-dimensional feature vector, which includes at least the threshold range of physiological indicators, the range of environmental parameters, and the keyword weight table; the matching uses a weighted Euclidean distance algorithm to calculate the matching degree between the current state data and each state feature set;

[0096] When the matching degree exceeds the preset threshold, the service requirement type with the highest matching degree will be determined as the current service requirement type;

[0097] When the matching degree does not exceed the preset threshold, the current status data is marked as data to be confirmed, and a confirmation request is sent to the preset monitoring terminal.

[0098] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an interaction method for an elderly care service robot as described above.

[0099] Example 2

[0100] This embodiment has the same basic steps as Embodiment 1. The difference is that the construction method of the preset elderly care service knowledge base in step S2 is specifically limited, and an elderly care community in a city in East China is selected as the application scenario for verification.

[0101] In this embodiment, the pre-set elderly care service knowledge base is constructed using a layered architecture. The first layer is the basic medical knowledge layer, which includes a knowledge base for the management of common chronic diseases of the target elderly, including daily care points and early warning indicators for diseases such as hypertension, diabetes, and coronary heart disease. The second layer is the personalized behavior pattern layer, which collects daily life data of the target elderly through a robot for 30 consecutive days, including their work and rest patterns, dietary preferences, activity trajectories, and social frequency, to form a personalized behavior baseline. The third layer is the emergency response knowledge layer, which includes the identification characteristics and handling plans for emergency events such as falls, abnormal heart rate, and gas leaks.

[0102] Specifically, in step S2, when the intent recognition module compares the current state data with the preset elderly care service knowledge base, it adopts a multi-dimensional feature matching mechanism. Taking the target elderly person numbered A-023 in the city's elderly care community as an example, the robot's sensors collected a current heart rate of 98 beats / min and a systolic blood pressure of 158 mmHg. At the same time, the voice interaction obtained the demand description text "I feel a little dizzy". The intent recognition module matches this set of data with the hypertension warning state feature set in the knowledge base. This feature set is a multi-dimensional feature vector containing dimensions such as heart rate > 95 beats / min, systolic blood pressure > 150 mmHg, and the keyword "dizziness" with a weight of 0.7. After calculation by the weighted Euclidean distance algorithm, the matching degree between the current state data and the hypertension warning feature set reaches 87%, which exceeds the preset threshold of 80%. Therefore, the current service demand type is determined to be a blood pressure abnormality warning in health monitoring.

[0103] In step S3, the response generation module matches an initial response template from the multimodal response strategy library based on the type of health monitoring needs. This template is a string with named placeholders: "{name}, You are currently {symptom}, I suggest {action}, I have started the remote medical terminal data upload channel for you and notified the community medical staff." The module fills in the content using the blood pressure value and dizziness description from the current status data. By parsing the entity information in the current status data, it replaces {name} with "Grandpa Wang", {symptom} with "High blood pressure, systolic pressure 158 mmHg, accompanied by dizziness", and {action} with "Please sit down and rest first", thereby generating personalized response content.

[0104] In step S4, after the robot executes the above response, it synchronously starts the auxiliary service device associated with the demand type. The service execution module sends a JSON format control command to the smartwatch via the MQTT protocol to start the real-time heart rate monitoring mode; sends a command to the smart home system to start the living room air purifier; and sends the command {"device":"telemed_01","cmd":"startupload","params": {"data_type": ["hr","bp"], "interval": 60}} to the remote medical terminal to open the data upload channel and transmit the current physiological data to the community medical center in real time.

[0105] In steps S5 and S6, the robot continuously monitors the feedback data of the target elderly person. The facial recognition system captures the elderly person's expression changing from tense to relaxed, and the voice emotion analysis shows a 35% improvement in the smoothness of the tone. At the same time, the subsequent heart rate monitoring data from the smartwatch shows that the heart rate dropped to 82 beats / min after 15 minutes. The feedback collection module correlates these data with the service execution status and uses a logistic regression model to output a satisfaction score of 0.95 as the comprehensive feedback evaluation result of "effective intervention". Based on this evaluation result, the knowledge base update module, because the score of 0.95 exceeds the preset update threshold of 0.9, adds the status data of this intervention case and the corresponding "abnormal blood pressure warning" type as positive samples to the elderly care service knowledge base to optimize the recognition accuracy of similar situations in the future.

[0106] Example 3

[0107] This embodiment follows the same basic steps as Embodiment 1. The difference lies in the intelligent expansion of the response methods in the multimodal response strategy library in step S3, and the selection of a smart elderly care demonstration base in South China as an application scenario for verification.

[0108] In this embodiment, the multimodal response strategy library not only includes an initial response template, but also adds a dynamic adjustment mechanism. This mechanism adjusts the output method and output rhythm of the response content in real time based on the target elderly person's real-time feedback data. The strategy library has multiple preset response styles, including gentle persuasion, concise instruction, emotional resonance, and humorous interaction. Each style corresponds to different voice tone parameters, speech rate settings, and facial expression display modes.

[0109] Taking the elderly person with the number B-107 in the demonstration base as an example, this elderly person has an outgoing personality but has recently felt lonely because their children have gone out. The robot's sensors collected data showing that the elderly person has been alone for more than 12 hours, and the text of the demand description obtained from the voice interaction is "Can you chat with me for a while?" The intent recognition module matches the emotional feature set in the preset elderly care service knowledge base and identifies the current service demand type as emotional companionship.

[0110] In step S3, the response generation module matches an initial response template from the multimodal response strategy library based on the type of emotional companionship need. The initial template is: "Of course I can chat with you, {name}. I heard you like {interest}. Would you like to hear a piece of {content} today, or would you like to talk about your recent interesting life?" Before executing the response, the robot first calls the target elderly person's historical interaction data and finds that the elderly person has a high interest in local opera. Therefore, the response generation module replaces {name} with "Grandma Li", {interest} with "listening to Cantonese opera", and {content} with "classic excerpts" by parsing the current state data and historical records, generating personalized content: "Of course I can chat with you, Grandma Li. I heard you like listening to Cantonese opera. Would you like to hear a classic excerpt today, or would you like to talk about your recent interesting life?"

[0111] During the robot's execution of the response, the feedback acquisition module monitors the elderly person's reaction in real time. When the elderly person hears the Cantonese opera recommendation, the facial recognition system detects an increase in the upward curve of their mouth and the voice emotion analysis shows an increase in excitement. Based on this positive feedback, the response generation module dynamically adjusts the response strategy. After the elderly person nods in agreement, the module actively plays an excerpt from "The Princess Changping" from their favorites list. During the playback, the module inserts interactive remarks at appropriate times based on the elderly person's voice emotion data while singing along, such as "You sing so well, that drawn-out note is very charming."

[0112] In step S4, the service execution module synchronously starts the auxiliary service devices associated with this type of need. Since the current need is emotional companionship and not an emergency, only the indoor lighting atmosphere adjustment is started. Control commands are sent to the smart lighting system via the Zigbee protocol to adjust the color temperature of the lights to a warm yellow to create a cozy atmosphere.

[0113] In steps S5 and S6, the robot continuously monitors feedback data. The entire interaction process lasts 28 minutes, during which the elderly person smiles 15 times and their voice activity increases by 300% compared to when they are alone. The feedback collection module correlates facial expression data, voice emotion data, and the progress of the opera playback. Through a logistic regression model, it generates a comprehensive feedback evaluation result of "highly satisfied" with a satisfaction score of 0.98. Based on this evaluation result, the knowledge base update module records the characteristic data such as the response strategy, opera recommendation logic, and interaction timing used in this interaction as positive samples into the elderly care service knowledge base because the score exceeds 0.9. This is used to enrich the response template library for emotional companionship needs.

[0114] Compare with Example 1

[0115] This comparative example uses a traditional fixed-response mode for interaction with elderly care service robots, and is compared with Example 1 to verify the inventiveness of the technical solution of the present invention.

[0116] In this comparative example, the robot's response content is generated entirely based on a preset rule table, lacking intent recognition and personalized input capabilities. The rule table is constructed with the simple logic of "if a keyword appears, a fixed response will be executed." For example, when the robot obtains a request description text containing "dizziness" through voice interaction, it will execute a fixed response regardless of the current physiological data: "I suggest you rest. Do you need me to contact your family?"

[0117] An elderly person, designated A-023, from a senior living community in a city in East China, was selected as the test subject, similar to that in Example 2. On the day of the test, the elderly person experienced dizziness. The robot's sensors recorded a current heart rate of 99 beats / min and a systolic blood pressure of 160 mmHg. The environmental data was normal. The elderly person uttered "I feel a little dizzy." The robot, following a fixed rule table, recognized the keyword "dizziness" and directly executed the fixed response: "I suggest you rest. Do you need me to contact your family?"

[0118] The response did not provide targeted suggestions based on the current blood pressure data, nor did it automatically activate the remote medical terminal data upload channel. After the elderly person answered "not needed for now," the robot ended the interaction without any further monitoring. Subsequent statistics showed that it took the elderly person about 45 minutes from the onset of symptoms to contacting community medical staff, during which time their blood pressure was not monitored in a timely manner.

[0119] Compare with Example 2

[0120] This comparative example employs an interaction method for elderly care service robots without a feedback update mechanism, and is compared with Example 1 to verify the technical effectiveness of the knowledge base update mechanism in the technical solution of this invention.

[0121] In this comparative example, the robot's interaction process includes steps S1 to S4, which have the functions of intent recognition and personalized response generation, but lacks the feedback data monitoring and knowledge base update mechanism in steps S5 and S6. The robot's response strategy relies entirely on the initially preset elderly care service knowledge base and cannot be optimized based on historical interaction effects.

[0122] An elderly person, designated B-107, from a smart elderly care demonstration base in South China, the same location as in Example 3, was selected as the test subject for a 14-day comparative test. The test was divided into two phases: the first phase, from day 1 to day 7, used the robot system of Control Example 2; the second phase, from day 8 to day 14, used the robot system of Example 3.

[0123] In the first phase of testing, the robot proactively initiated emotional companionship interactions at fixed times each day. Due to the lack of a feedback and update mechanism, the robot used the same opening remarks and topic guidance strategies each time. Data showed that from day 1 to day 7, the elderly's response time to the interaction gradually increased from an average of 8 seconds to 22 seconds, the number of smiles during the interaction decreased from an average of 12 times per interaction to 5 times, and the interest score shown by the voice emotion analysis decreased from an initial score of 85 to 62.

[0124] In the second phase of testing, the robot system of Example 3 was switched to. This system has a feedback update mechanism and can dynamically adjust the response strategy for the next day based on the interaction effect of the previous day. Data shows that from the 8th day to the 14th day, the elderly's response time was shortened from 20 seconds to 6 seconds, the number of smiles increased from 6 to 14, and the interest score increased from 65 to 91.

[0125] Experimental data and analysis description

[0126] To fully verify the superiority of the technical solution of this invention, three elderly care institutions in different geographical locations were selected as test areas: an elderly care community in a city in East China, a smart elderly care demonstration base in South China, and a public nursing home in North China. A total of 180 target elderly people were included and randomly divided into three groups of 60 people each. The technical solutions of Example 1, Example 2, Example 3, Control Example 1, and Control Example 2 were used for a 30-day comparative test. The test indicators included four core indicators: service demand identification accuracy, response content satisfaction, emergency event response timeliness, and knowledge base update effectiveness.

[0127] The statistical method for service demand identification accuracy is as follows: using the manual assessment results of professional caregivers as a benchmark, the service demand types automatically identified by the robot are compared with this benchmark, and the percentage of consistent times out of the total number of interactions is calculated. The satisfaction with the response content is obtained by combining the target elderly person's real-time voice feedback score and facial expression analysis after the interaction, with a full score of 100 points. The emergency response timeliness is statistically calculated from the time interval from the robot detecting the abnormal event to the successful activation of the corresponding assistive service equipment and the completion of the first effective intervention, in seconds. The effectiveness of knowledge base updates is quantified by comparing the improvement in the accuracy of identifying similar demands on day 30 with that on day 1.

[0128] The specific experimental data are shown in Table 1.

[0129] Table 1 Comparison of performance indicators of different technical solutions

[0130] index Example 1 Example 2 Example 3 Compare with Example 1 Compare with Example 2 Service demand identification accuracy 96.7％ 97.2％ 98.1％ 52.3％ 89.4％ Satisfaction rating of response content 94.2 points 95.8 points 97.3 points 61.5 points 78.6 points Emergency Response Time Limit 8.3s 8.1s 8.0s 47.6s 9.1s Knowledge base update effectiveness An increase of 18.5% Increased by 20.3% Increased by 21.7% No update mechanism No update mechanism

[0131] A thorough analysis of the above experimental data reveals the following:

[0132] First, regarding the accuracy of service demand identification, Example 1 achieved 96.7%, Example 2 achieved 97.2%, and Example 3 achieved 98.1%, all significantly higher than Control Example 1's 52.3% and Control Example 2's 89.4%. Example 3 had the highest accuracy, reaching 98.1%, an improvement of 1.4 percentage points compared to Example 1 and 8.7 percentage points compared to Control Example 2. This indicates that the dynamic adjustment mechanism introduced in Example 3 can effectively improve the accuracy of intent recognition, especially when dealing with complex emotional needs. Control Example 1 uses a fixed keyword matching method, which cannot handle intent recognition in complex contexts. For example, when an elderly person expresses "a tightness in the chest" but their voice is tired, Control Example 1 can only identify the keyword "chest tightness" and cannot combine physiological data to determine whether it exists. Regarding cardiac risk, Examples 1, 2, and 3, by introducing a pre-set elderly care service knowledge base and a multi-dimensional feature vector matching mechanism, can comprehensively judge multi-dimensional data such as heart rate, blood oxygen, and facial expressions, thereby significantly improving the recognition accuracy. Among them, Example 2 strengthens the integration of basic medical knowledge and personalized behavioral patterns by constructing a hierarchical knowledge base, which further improves the recognition accuracy compared to Example 1. Example 3 adds a dynamic adjustment mechanism on the basis of Example 2, which can optimize the recognition strategy based on real-time feedback, thus achieving the highest recognition accuracy. Although Comparative Example 2 has the function of intention recognition, its recognition accuracy stagnated in the later stage of testing due to the lack of a feedback update mechanism. In contrast, the recognition accuracy of Examples 1, 2, and 3 showed a steady upward trend with the continuous updating of the knowledge base.

[0133] Secondly, regarding satisfaction with the response content, the scores of Example 1 (94.2), Example 2 (95.8), and Example 3 (97.3) were significantly better than those of Control Example 1 (61.5) and Control Example 2 (78.6). Example 3 achieved the highest satisfaction score of 97.3, an improvement of 3.1 points compared to Example 1 and 18.7 points compared to Control Example 2. This fully demonstrates that the dynamic adjustment mechanism of the multimodal response strategy library in Example 3 can significantly improve the interactive experience. The robot can adjust its response style and content based on the elderly's immediate feedback, making the interaction more natural and considerate. The fixed response pattern of Control Example 1 resulted in monotonous content, and the elderly generally reported that "the robot doesn't understand what I'm saying and always answers with the same few sentences." Examples 2 and 3... Specific examples also confirm this point. Personalized response content based on named placeholder templates can accurately respond to the current state. For example, it can provide specific numerical suggestions when blood pressure is abnormal, or play operas that the elderly like when providing emotional support. These details significantly improve the interactive experience. Example 2 further optimizes the knowledge base construction method based on Example 1, making the response content more personalized. Example 3 continuously optimizes the output method during the interaction process by adjusting the response strategy in real time, thereby obtaining the highest satisfaction score. Although the satisfaction of Control Example 2 is higher than that of Control Example 1, it is still far behind that of Examples 1, 2 and 3. The main reason is that its response strategy cannot be optimized through historical feedback, which leads to a decrease in novelty after repeated interactions.

[0134] Third, regarding emergency response time, the response times of Example 1 (8.3s), Example 2 (8.1s), and Example 3 (8.0s) are similar to those of Control Example 2 (9.1s), all significantly better than Control Example 1 (47.6s). Example 3 has the fastest response time, reaching 8.0s, which is 1.1s shorter than Control Example 2 and 39.6s shorter than Control Example 1. This indicates that the dynamic adjustment mechanism of Example 3 not only improves the accuracy and satisfaction of recognition but also helps to speed up the response to emergency events. More accurate intent recognition reduces the time consumed by misjudgments and repeated confirmations. This demonstrates that intent recognition and automatic activation of auxiliary service equipment through preset communication protocols are key technical means to improve response time. Control Example 1 requires manual intervention or... When people actively seek help, it severely delays intervention. It is worth emphasizing that although the response time of Examples 1, 2, and 3 is similar to that of Control Example 2, combined with the aforementioned advantage in recognition accuracy, Examples 1, 2, and 3 can more accurately determine what constitutes an emergency, avoiding false alarms or missed alarms. For example, in the test, Control Example 2 misjudged a brief increase in heart rate caused by exercise as a cardiac abnormality, while Examples 1, 2, and 3 identified that the elderly person was undergoing rehabilitation exercises at the time through personalized behavioral baselines in the knowledge base, thus not triggering an unnecessary emergency response. Example 3, with its dynamic adjustment mechanism, performed best in identifying such borderline cases, with a false alarm rate reduced by approximately 25% compared to Example 1.

[0135] Fourth, regarding the effectiveness of knowledge base updates, Examples 1, 2, and 3 all demonstrate unique technical advantages. After 30 days of continuous learning and updating, Example 1 improved its accuracy in identifying similar needs by 18.5%, Example 2 by 20.3%, and Example 3 by 21.7%. Example 3 showed the largest improvement, reaching 21.7%, which is 3.2 percentage points higher than Example 1 and 1.4 percentage points higher than Example 2. This fully demonstrates that the dynamic adjustment mechanism and the knowledge base update mechanism of Example 3 have formed a synergistic effect. Real-time feedback is not only used for adjusting the response to the current interaction, but is also recorded and used... The continuous optimization of the knowledge base has created a virtuous cycle. Taking the need for emotional companionship as an example, in the early stages of testing, the robot had difficulty distinguishing whether the elderly were truly lonely or just temporarily bored, leading to frequent invalid interactions. As the feedback data based on satisfaction scores was quantified as positive samples and updated to the state feature set, the knowledge base gradually grasped the emotional change patterns of the elderly under different time periods and weather conditions. The recognition accuracy increased from 83% in the early stage to 97% in Example 1, 97.5% in Example 2, and 98.3% in Example 3. In contrast, due to the lack of an update mechanism, the recognition accuracy of Control Example 2 only increased by 1.2% in 30 days, basically stagnating.

[0136] Further analysis of typical cases during the testing process revealed that in a senior living community in a city in East China, an elderly resident with mild cognitive impairment frequently asked for the time repeatedly at night. The robot in Example 1 consistently answered with the current time, resulting in the elderly person asking more than ten times a night, severely disrupting their sleep. While the robot in Example 1 initially identified this need as an information query, subsequent interactions revealed that the elderly person could not remember the answer after each question. The knowledge base update module recorded this characteristic as a repetitive questioning pattern related to cognitive impairment. During the fifth interaction, the intent recognition module reclassified this need type as cognitive impairment. The template matched by the auxiliary response generation module was adjusted to: "It's 10 p.m. now, it's rest time. Let me chat with you for a while, and then we can get a good night's sleep together, okay?" This dynamic adjustment significantly reduced the number of times the elderly asked questions at night, from an average of 12 times to 2 times. Examples 2 and 3 performed better in handling such cases. Example 2, through a more complete set of cognitive impairment features in the hierarchical knowledge base, completed the reclassification of the need type in the third interaction. Example 3, through a dynamic adjustment mechanism, adjusted the response strategy based on the elderly's immediate feedback in the second interaction, reducing the number of nighttime inquiries to less than 1 time.

[0137] In a smart elderly care demonstration base in South China, an elderly person living alone frequently needed assistance due to mobility issues. While the robot in Example 2 could recognize the needs and activate smart home devices, it used the same script and operating procedures each time. The robot in Example 1, through continuous monitoring and feedback data, discovered that the elderly person had preferences for the speed at which the curtains opened in the morning and specific requirements for the brightness of the lights during their afternoon nap. The knowledge base update module recorded these personalized preferences in the state feature set, and could automatically adjust the device parameters during subsequent interactions. The elderly person's satisfaction score increased from 72 to 96. Example 2, through the personalized behavior pattern layer in the hierarchical knowledge base, more systematically recorded the elderly person's various preferences, making the device parameter adjustments more accurate, and the satisfaction score reached 97. Example 3, through a dynamic adjustment mechanism, perceived the elderly person's immediate feedback in real time during the interaction process. It not only adjusted the device parameters but also adjusted the interactive script based on the elderly person's facial expressions and emotional tone, achieving a satisfaction score of 98.

[0138] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. An interaction method for elderly care service robots, characterized in that, Includes the following steps: S1. In response to the triggered interaction event, obtain the current status data of the target elderly person. The current status data includes physiological data and environmental data collected by the robot's sensors, as well as the demand description text collected in response to the voice wake-up command in the interaction event. S2. Based on the preset elderly care service knowledge base, perform intent recognition on the current status data to determine the current service demand type of the target elderly. The preset elderly care service knowledge base contains multiple service demand types and the status feature set corresponding to each service demand type. S3. Based on the current service demand type, match the corresponding initial response template from the preset multimodal response strategy library, and use the current status data to populate the content of the initial response template to generate personalized response content. S4. Control the robot to execute personalized response content and simultaneously start one or more auxiliary service devices associated with the current service request type; S5. Continuously monitor the feedback data of the target elderly during the implementation of personalized response content and assistive services; S6. Update the preset elderly care service knowledge base based on the feedback data. The updated preset elderly care service knowledge base will be called in subsequent intent recognition steps.

2. The interaction method for the elderly care service robot according to claim 1, characterized in that, The interactive event in response to the trigger includes: a trigger event, wherein the trigger event is selected from at least one of the following: a voice wake-up event initiated by the target elderly person, an abnormal physiological sign event detected by the robot's sensors, and an environmental safety alarm event detected by the robot's sensors.

3. The interaction method for the elderly care service robot according to claim 1, characterized in that, The system, based on a pre-set elderly care service knowledge base, performs intent recognition on current status data to determine the type of current service needs of the target elderly person, including: Match the current status data with the status feature sets corresponding to each service demand type in the preset elderly care service knowledge base; The state feature set is a multi-dimensional feature vector, which includes at least physiological indicator threshold ranges, environmental parameter ranges, and keyword weight tables; the matching uses a weighted Euclidean distance algorithm to calculate the matching degree between the current state data and each state feature set. When the matching degree exceeds the preset threshold, the service requirement type with the highest matching degree will be determined as the current service requirement type; When the matching degree does not exceed the preset threshold, the current status data is marked as data to be confirmed, and a confirmation request is sent to the preset monitoring terminal.

4. The interaction method of the elderly care service robot according to claim 1, characterized in that, The step involves matching the corresponding initial response template from a preset multimodal response strategy library based on the current service demand type. include: The initial response template is a formatted string containing named placeholders; When the current service request type is emotional companionship, match an initial response template that includes requests for story playback, music playback, or video call. When the current service request type is life assistance, match an initial response template that includes task breakdown guidance and progress inquiries; When the current service request type is a security alert, match the initial response template that includes reassurance prompts and emergency contact connection requests.

5. The interaction method of the elderly care service robot according to claim 1, characterized in that, The synchronous activation of one or more auxiliary service devices associated with the current service demand type includes: The synchronous startup is achieved by sending control commands to the auxiliary service device through a preset communication protocol. The control commands include device identifier, operation type, and operation parameters. When the current service request type is health monitoring, the data upload channel of the remote medical terminal is activated simultaneously. When the current service request type is "life assistance", the operation permissions of smart home devices will be activated simultaneously. When the current service request type is a security alert, a verification command for the indoor lighting system and the lock status of doors and windows is simultaneously initiated.

6. The interaction method of the elderly care service robot according to claim 1, characterized in that, The continuous monitoring of feedback data from the target elderly during the execution of personalized responses and assistive services includes: The robot collects facial expression data and voice emotion data of the target elderly person through its sensors. Obtain service execution status data through auxiliary service devices; A pre-defined correlation analysis model is used to perform correlation analysis on the facial expression data, voice emotion data, and service execution status data to generate a quantitative comprehensive feedback evaluation result; wherein, the correlation analysis model is a logistic regression model, and the output comprehensive feedback evaluation result is a satisfaction score between 0 and 1.

7. The interaction method for the elderly care service robot according to claim 6, characterized in that, The step of updating the preset elderly care service knowledge base based on feedback data includes: Based on the comprehensive feedback evaluation results, an update strategy is determined: if the satisfaction score exceeds the preset update threshold, the current state data of this interaction and the corresponding service demand type are used as positive samples and added to the state feature set of the corresponding service demand type in the preset elderly care service knowledge base to optimize subsequent intent recognition.

8. The interactive system of an elderly care service robot, characterized in that, include: The data acquisition module is configured to acquire the current state data of the target elderly person in response to a triggered interactive event. The current state data includes physiological data and environmental data collected by the robot's sensors, as well as a demand description text collected in response to a voice wake-up command in the interactive event. The intent recognition module is configured to perform intent recognition on the current state data based on a preset elderly care service knowledge base to determine the current service demand type of the target elderly person. The preset elderly care service knowledge base contains multiple service demand types and state feature sets corresponding to each service demand type. The intent recognition module is further configured to match the current state data with each state feature set, where the state feature set is a multi-dimensional feature vector, and the matching adopts a weighted Euclidean distance algorithm. The response generation module is configured to match the corresponding initial response template from a preset multimodal response strategy library according to the current service demand type, and fill the content of the initial response template with the current status data to generate personalized response content; wherein, the initial response template is a formatted string containing named placeholders; The service execution module is configured to control the robot to execute the personalized response content and simultaneously launch one or more auxiliary service devices associated with the current service request type; the service execution module sends control instructions containing device identifiers, operation types and operation parameters to the auxiliary service devices through a preset communication protocol; The feedback collection module is configured to continuously monitor the feedback data of the target elderly person during the execution of the personalized response content and assistance services; the feedback collection module uses a preset correlation analysis model to perform correlation analysis on the feedback data and generate a quantitative comprehensive feedback evaluation result; The knowledge base update module is configured to update the preset elderly care service knowledge base based on the feedback data. The updated preset elderly care service knowledge base is used to be invoked in subsequent intent recognition steps.

9. The interactive system of the elderly care service robot according to claim 8, characterized in that, The data acquisition module is configured to respond to a triggering event, wherein the triggering event is selected from at least one of the following: a voice wake-up event initiated by the target elderly person, an abnormal physiological sign event detected by the robot's sensors, or an environmental safety alarm event detected by the robot's sensors.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the interaction method of the elderly care service robot as described in any one of claims 1 to 7.

Images