Intelligent accompanying robot based on ai and internet of things and nursing cooperation system and method

Abstract

The application discloses an intelligent accompanying and nursing robot and a nursing cooperation system and method based on AI and the Internet of Things, belongs to the technical field of robots, and solves the problem that the adaptability, intelligent level and precision of man-machine cooperation of the system are insufficient in a complex nursing scene based on the fact that a control module determines different control modes of a nursing robot according to input information; the system comprises an Internet of Things sensing module, a decision control module, a motion execution module and an intelligent nursing robot; in the application, the Internet of Things sensing module collects multi-dimensional data such as action posture, facial expression, voice instruction and contact pressure of a nursing object in real time, and then the AI intention recognition model in the decision control module performs deep analysis, so that the real nursing demand of the old person can be accurately inferred, the corresponding nursing strategy can be automatically generated according to the intention recognition result, manual intervention is not needed, and the adaptability, safety and service precision of the intelligent accompanying and nursing robot in a health care scene are improved.

Description

Technical Field

[0001] This invention belongs to the field of robotics technology, specifically relating to intelligent companion robots and nursing collaboration systems and methods based on AI and the Internet of Things. Background Technology

[0002] With the increasing aging of the population and the growing strain on medical resources, traditional elderly care models face numerous challenges. In places such as nursing homes and hospitals, the number of nursing staff is relatively insufficient, and a lot of time and energy needs to be spent on the daily care of patients or the elderly, such as assisting with turning over, wound care, perineal cleaning and other basic life care tasks.

[0003] Traditional manual care has many inconveniences. Caregivers need to be on duty for long periods, which can easily lead to fatigue and inconsistent quality of care. Furthermore, wound care, such as cleaning wounds and perineal hygiene, requires specific professional knowledge and skills. In the absence of sufficient caregivers, these tasks may not be completed in a timely and standardized manner, impacting the patient's recovery process.

[0004] In recent years, robotics technology has gradually emerged in the fields of medical care and elderly care. Robots can work tirelessly and accurately complete various tasks according to preset programs. However, the application of robots in elderly care is still not perfect. Most robots can only perform single functions, such as assisting in rehabilitation training or performing simple cleaning tasks, lacking comprehensive nursing collaboration capabilities.

[0005] Chinese patent CN118700169A discloses an intelligent elderly care robot system and control method, including a care robot equipped with a visual perception module, a force perception module, and a control module. The control module determines different control modes for the care robot based on input information, and controls the care robot to move to a target location to grasp a target object according to different control modes and control methods. However, existing systems rely on the control module to determine different control modes for the care robot based on input information. The decision-making logic depends on preset rules rather than actively understanding the elderly person's intentions, resulting in weak intention understanding ability and insufficient adaptability, intelligence level, and accuracy of human-machine collaboration in complex care scenarios. To address the above problems, we propose an intelligent companion robot and nursing collaboration system and method based on AI and IoT. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing an intelligent companion robot and nursing collaboration system and method based on AI and IoT. This invention solves the problems of existing systems that determine different control modes for nursing robots based on input information by the control module, and whose decision-making logic relies on preset rules rather than actively understanding the elderly's intentions. This results in weak intention understanding capabilities and insufficient adaptability, intelligence level, and accuracy of human-machine collaboration in complex nursing scenarios.

[0007] This invention is implemented as follows: an intelligent companion robot and nursing collaboration system based on AI and IoT, comprising:

[0008] The Internet of Things (IoT) sensing module is used to acquire monitoring sensor data of the nursing environment, preprocess the monitoring sensor data, and obtain a real-time dataset.

[0009] The decision control module is used to pre-build an AI-based intent recognition model, iteratively train the intent recognition model using historical datasets, output a converged intent recognition model, acquire real-time datasets, identify and analyze the real-time datasets based on the intent recognition model, generate intent recognition results, generate dynamic planning paths based on dynamic window path planning method combined with intent recognition results, and trigger nursing collaboration strategies in real time using dynamic planning paths as indexes.

[0010] The action execution module is used to acquire nursing collaboration strategies, parse and identify the nursing collaboration strategies, and generate action execution instructions;

[0011] Intelligent nursing robots are used to respond to the action execution commands and perform nursing collaborative work.

[0012] Preferably, the intelligent nursing robot includes:

[0013] The robot body includes a main body base, a lifting platform, and a nursing workbench. The lifting platform is mounted on the main body base, and the top of the lifting platform is detachably connected to the nursing workbench.

[0014] At least one set of adjustable nursing assistance arms, said nursing assistance arms being mounted on both sides of the nursing workbench;

[0015] A nursing transport mechanism that works in conjunction with a nursing assistance arm. The nursing transport mechanism is located inside the nursing workstation and is used for the automatic transfer and delivery of nursing supplies.

[0016] Preferably, the nursing assistance arm includes:

[0017] The first swing motor is fixedly installed inside the nursing workbench;

[0018] The first swing arm is fixedly connected to the output shaft of the first swing motor;

[0019] The second swing motor is fixedly installed at the end of the first swing arm, and the output shaft of the second swing motor is fixedly connected to the second swing arm;

[0020] A third swing motor is fixedly installed at the end of the second swing arm, and the output shaft of the third swing motor is fixedly connected to the first cylinder;

[0021] An adaptive clamping part is fixedly connected to the telescopic rod of the first cylinder, which is used to adaptively clamp and release care items.

[0022] Preferably, the adaptive clamping part includes:

[0023] The clamping mounting base is detachably mounted on the end of the telescopic rod of the first cylinder, and at least one set of clamping positioning grooves is provided inside the clamping mounting base;

[0024] The first adjusting motor is fixedly installed at the end of the clamping and positioning groove;

[0025] Rotate the adjusting screw installed in the clamping and positioning groove; one end of the adjusting screw is fixedly connected to the adjusting motor.

[0026] An adaptive clamping seat is threaded onto the outside of the adjusting screw, and the adaptive clamping seat is slidably connected to the clamping and positioning groove.

[0027] Preferably, the nursing transport mechanism includes:

[0028] The second cylinder is fixedly installed on the surface of the nursing workbench;

[0029] A T-shaped seat is fixedly connected to the telescopic rod of the second cylinder, and the T-shaped seat is slidably connected to the nursing workbench;

[0030] The second adjusting motor is fixedly installed on the side wall of the T-shaped seat;

[0031] The transfer linkage unit connected to the second regulating motor, and

[0032] The transfer and adsorption section is connected to the transfer linkage section and is used to adsorb care items.

[0033] Preferably, the transfer linkage unit includes:

[0034] A first hinge arm is fixedly connected to the output shaft of a second regulating motor, and a third swing arm is provided on the side of the first hinge arm away from the second regulating motor.

[0035] The third swing arm is rotatably mounted on the T-shaped seat, and the third swing arm has an upper hinge groove and a lower hinge groove. One end of the first hinge arm is hinged to the lower hinge groove.

[0036] The second hinge arm is located on one side of the third swing arm, and one end of the second hinge arm is hinged to the upper hinge groove. The other end of the second hinge arm passes through the vertical stand and is fixedly connected to a swing hinge sleeve.

[0037] The first slide rod is slidably installed inside the swing hinge sleeve, and one end of the first slide rod is fixedly connected to the hinge sleeve connecting seat;

[0038] A vibration damping and protection spring is sleeved on the outside of the first slide rod, and one end of the vibration damping and protection spring is fixedly connected to the swing hinge sleeve;

[0039] A transfer moving part is detachably connected to the hinged sleeve connecting seat. The transfer moving part includes a transfer moving seat, a second slide rod, a third slide rod, and a connecting slide sleeve. The transfer moving seat is slidably sleeved on the second slide rod. The side wall of the transfer moving seat is rotatably connected to the hinged sleeve connecting seat. One end of the rotatable moving seat is detachably connected to a transfer adsorption part. The end of the second slide rod is fixedly connected to a connecting slide sleeve. The connecting slide sleeve is slidably sleeved on the third slide rod. The third slide rod is detachably installed in the T-shaped seat.

[0040] The transport and adsorption section includes:

[0041] A transfer adsorption seat, which is detachably mounted on the end of a transfer movable seat;

[0042] At least one set of adsorption negative pressure tanks are provided, the adsorption negative pressure tanks are opened in the transfer adsorption seat, and a negative pressure pump is provided in the transfer adsorption seat, the negative pressure pump is connected to the adsorption negative pressure tanks.

[0043] Preferably, the IoT sensing module includes:

[0044] The displacement sensor, installed inside the main body base, is used to sense the robot's displacement changes in real time, convert the displacement data into electrical signals and transmit them.

[0045] Visual sensors are used to collect image information of the person being cared for, in order to identify the person's movements, facial expressions, skin condition, as well as the position and state of objects in the surrounding environment, and generate visual monitoring data.

[0046] A voice sensor, installed inside the nursing workstation, is used to collect the voice commands and voice interaction information of the person being cared for in real time;

[0047] Pressure sensors are used to collect the contact pressure between the object being cared for in real time and convert the pressure data into electrical signals for transmission.

[0048] The preprocessing unit is communicatively connected to the displacement sensor, vision sensor, voice sensor, and pressure sensor. It is used to receive displacement data, visual monitoring data, voice commands and voice interaction information, and pressure data, and integrate them into monitoring sensor data. It performs normalization and data cleaning preprocessing on the monitoring sensor data and outputs a real-time dataset.

[0049] The adaptive clamping holder uses a bidirectional coupled anti-resistance control algorithm to adjust the operating parameters of the clamping holder motor in real time. First, based on monitoring sensor data, and considering the combination of manual and visual methods, a receiver intention coefficient is defined. , ,in express The cumulative duration of the caregiver's gaze at the clamping seat is displayed in the sliding window. Indicates the fixation duration threshold. Indicates the speed at which the person being cared for approaches with their hand. This indicates the maximum permissible hand approach speed. The adjustment coefficient is then used. Next, a weighted integral is performed on the rate of change in the gripping force of the person being cared for to quantify the smoothness of the interaction between the person being cared for and the gripper, defining the contact disturbance coefficient. This characterizes the smoothness of the interaction between the person being cared for and the clamping device. ,in This indicates the rate of change of contact force of the clamping seat pressure sensor. Indicates the time forgetting factor, Represents the integral variable. The stability coefficient is used to identify smooth contact force as an effective contact force signal, and high-frequency oscillating force as an indicator that the person being cared for has not effectively grasped the item, thus serving as an invalid contact force signal. Next, the control stiffness of the gripper motor is defined using the receiver's intention coefficient and the contact disturbance coefficient. Stiffness indicates the clamping force of the clamping seat motor. ,in, Indicates the clamping stability reference stiffness. This indicates that the person being cared for is exerting an outward pulling force. Indicates standard interaction force, This indicates the minimum contact stiffness at which the clamping mechanism is held. This represents the activation function. This represents the disturbance sensitivity factor. Finally, based on the position increment integral structure and the contact force deviation and motor control stiffness, the position is updated according to the force state. The position update formula is: ,in Indicates the current position of the motor. Indicates the motor position after the command is executed. This represents the residual stress under interactive conditions. Indicates the position response parameters. This represents the integral variable, and the position is controlled by the motor to achieve a smooth interaction between the clamp and the items being cared for.

[0050] The decision control module includes:

[0051] The model training unit is used to pre-build an AI-based intent recognition model, iteratively train the intent recognition model using historical datasets, and output a converged intent recognition model.

[0052] The intent recognition model is based on a convolutional neural network (CNN) model and includes an input layer and an output layer. A multimodal encoder is set between the input layer and the CNN model, and an adaptive filter is set between the multimodal encoder and the input layer. An attention mechanism is introduced in the multimodal encoder, which includes two Conv2D layers, a Transformer encoding layer, a fully connected layer, and an average pooling layer. A cross-modal attention mechanism is introduced in the CNN model. An LSTM layer is introduced between the CNN model and the output layer. The LSTM layer dynamically adjusts the weights of each modality feature based on a gating fusion mechanism and combines the Hidden Markov Model (HMM) algorithm to fuse and weight the modality features after weight adjustment to obtain multimodal fused weighted features. The intent comprehensive probability distribution of the multimodal fused weighted features is output based on the Softmax activation function.

[0053] The identification and analysis unit identifies and analyzes real-time datasets based on the intent recognition model, and generates intent recognition results.

[0054] The path planning unit generates a dynamic planning path based on the dynamic window path planning method combined with the intent recognition result, and uses the dynamic planning path as an index to trigger nursing collaboration strategies in real time.

[0055] Preferably, the method for identifying and analyzing real-time datasets based on an intent recognition model includes:

[0056] Acquire a real-time dataset, identify the data types in the real-time dataset, perform adaptive filtering on the real-time data based on the data types, and output the filtered dataset.

[0057] In adaptive filtering, the filtering formula is expressed as follows:

[0058]

[0059]

[0060] in, These are the filtered dataset and the original signal in the real-time dataset, respectively. These are the adaptive filter coefficients. For adaptive filtering time step, These are the filter coefficients from the previous time step. These are the step size parameter and the error signal, respectively.

[0061] Load the filtered dataset, extract the modal feature vectors of the multimodal data from the filtered dataset based on the autoencoder, and perform cross-modal alignment of the modal feature vectors based on the attention mechanism, and output the cross-modal aligned feature vector set;

[0062] The feature vector set is obtained, and the convolutional neural network model performs convolutional fusion processing on the feature vector set to extract the fused modal features.

[0063] The fused modal features are represented as follows:

[0064]

[0065]

[0066] in, These are the fused modal features and the first convolutional result of the convolutional neural network model, respectively. This is the input representation of the feature vector set. These are the weight matrices of the first and second convolutional kernels in the convolutional neural network model, respectively.

[0067] The weights of each modal feature are dynamically adjusted based on the gating fusion mechanism, and the weighted modal features after weight adjustment are fused and weighted by the Hidden Markov algorithm to obtain multimodal fusion weighted features.

[0068] The multimodal fusion weighted features are represented as follows:

[0069]

[0070] in, These are visual modal features, speech modal features, and sensory modal features, respectively. These are the weights for visual modal features, speech modal features, and sensor modal features, respectively.

[0071]

[0072]

[0073]

[0074] The doorway mechanism score is based on visual modality features. For the gate mechanism score of speech modality features, The gate mechanism score is assigned to the sensor modal characteristics.

[0075] Based on the Softmax activation function, the comprehensive probability distribution of intent is output using multimodal fusion weighted features. The intent recognition result is then output, which includes the target location and operation type.

[0076] Preferably, a dynamic task scheduling algorithm is proposed. Based on actual nursing scenarios, a time- and space-based demand potential field is constructed to calculate task priorities. First, based on data from voice sensors, pressure sensors, and vision sensors, the first... Comprehensive nursing entropy of each monitoring node , ,in, , , These represent the speech weighting coefficient, the pressure weighting coefficient, and the image weighting coefficient, respectively. Indicates the first Data from a voice sensor, This represents the mean of the background noise. Indicates the first Data from a pressure sensor, Indicates the minimum pressure threshold. This represents the activation function. Indicates the first Confidence level of abnormal behavior identified by a visual sensor The standard deviation of environmental noise is represented by the comprehensive nursing entropy. The degree of current environmental anomaly is quantified using this comprehensive nursing entropy. Then, based on this comprehensive nursing entropy, distance attenuation and time attenuation are introduced to construct a demand potential energy field, and the comprehensive nursing potential energy at the anomaly source is calculated. ,in, Indicating intelligent companion robots Time for the first Comprehensive nursing potential of each abnormal source Indicates the time decay coefficient. Indicates the first The trigger time of each abnormal source Indicates the first The distance between the anomaly source and the location of the intelligent companion robot Indicates the spatial decay index, Let the spatial error coefficient be represented. Then, the task complexity coefficient is defined. The task cost is calculated through task consumption and operational risk, and a cost function is constructed. ,in, This indicates that intelligent nursing robots have reached the [number]th [number]. The overall execution cost of a task based on anomaly sources. This indicates that intelligent nursing robots have reached the [number]th [number]. Calculate the path planning distance for the task based on the anomaly source. This indicates the battery percentage of the intelligent nursing robot. Indicates the first The task complexity coefficient of each abnormal source task , , This represents the cost weighting coefficient. Finally, based on the comprehensive nursing potential and comprehensive execution cost ratio, a nonlinear gain mechanism is used to amplify the task advantages, improve the sensitivity of nursing tasks, and construct a global evaluation index for nursing tasks. , ,in, This represents the total number of tasks, based on the overall evaluation index of nursing tasks. This reflects the importance of intelligent nursing robots for the first The probability of task activation for each abnormal source.

[0077] Preferably, the method for generating dynamic planning paths based on dynamic window path planning combined with intent recognition results includes:

[0078] The intent recognition result is obtained, and the target point spatial coordinates, operation type, type priority, and target point real-time environmental data are determined based on the intent recognition result. The target point spatial coordinates, operation type, type priority, and target point real-time environmental data are mapped to the state space, and the data in the state space is converted into a high-dimensional feature vector through the decoder.

[0079] Load high-dimensional feature vectors, capture the global dependency between the target point and local obstacle points through a self-attention mechanism, and gradually generate the initial path between the target point and the robot's current coordinates through a generative adversarial network;

[0080] Obtain the initial path between the current coordinate point and the target point. Then, adjust the path segments of the initial path by combining the global dependency relationship between the target point and local obstacle points using the dynamic window path planning method to obtain the adjusted path.

[0081] The system uses proximity to the target location, path smoothness, and priority as positive rewards, and collision risk and deviation from the target as negative penalties. It constructs a multi-objective reward function by combining the waypoints of the adjusted path, and solves the multi-objective reward function based on the near-end strategy optimization to generate a dynamic programming path.

[0082] The multi-objective reward function is expressed as:

[0083]

[0084] in, These are, respectively, the target proximity weight coefficient, the path smoothness weight coefficient, the task priority weight coefficient, the collision risk weight coefficient, and the deviation from the target weight coefficient. Each represents the current time. Robot's current coordinates, target coordinates, and previous time point. Robot coordinates, next moment Robot coordinates The set of obstacle points at the current moment; These are reward items for target proximity, path smoothness, task priority, collision risk, and deviation from the target.

[0085] On the other hand, the present invention also provides a method for intelligent companion robots and nursing collaboration based on AI and the Internet of Things, the method comprising:

[0086] Acquire nursing environment monitoring sensor data, preprocess the monitoring sensor data to obtain real-time dataset;

[0087] A pre-built AI-based intent recognition model is used to iteratively train the intent recognition model using historical datasets, and a converged intent recognition model is output.

[0088] Acquire real-time datasets, analyze and identify the real-time datasets based on intent recognition models, generate intent recognition results, generate dynamic planning paths based on dynamic window path planning methods combined with intent recognition results, and trigger nursing collaboration strategies in real time using dynamic planning paths as indexes.

[0089] The system acquires nursing collaboration strategies, analyzes and identifies these strategies, generates action execution instructions, and the intelligent nursing robot responds to these instructions to perform nursing collaboration tasks.

[0090] Compared with the prior art, the embodiments of this application have the following main advantages:

[0091] In this embodiment of the invention, the IoT sensing module collects multi-dimensional data such as the subject's movements, facial expressions, voice commands, and contact pressure in real time. Then, the AI ​​intent recognition model in the decision control module performs in-depth analysis to accurately infer the elderly person's true care needs. Based on the intent recognition results, corresponding care strategies can be automatically generated without human intervention, thus improving the adaptability, safety, and service accuracy of the intelligent companion robot in the health and wellness scenario.

[0092] In this embodiment of the invention, the intelligent nursing robot consists of a robot body, a nursing assistance arm, and a nursing transfer mechanism. The nursing assistance arm and the nursing transfer mechanism work together to meet the requirements of companionship and nursing collaboration in different scenarios. The nursing transfer mechanism is responsible for automatically transferring nursing supplies from the storage area to a designated location on the nursing workbench. The nursing assistance arm accurately delivers items to the elderly person or directly assists them in using them, avoiding the tediousness of frequent manual retrieval and solving the problems of slow item retrieval and inaccurate positioning in traditional nursing care. This improves the response speed of basic nursing care. Furthermore, the collaborative work between the nursing assistance arm and the nursing transfer mechanism allows the robot to adapt to various nursing scenarios and meet the needs of different care recipients. The robot can automatically adjust its operation mode and item delivery method according to the specific scenario, ensuring the efficiency and accuracy of nursing operations and improving the adaptability and flexibility of nursing care. Through multimodal perception and adaptive adjustment mechanisms, the comfort and stability of nursing operations are ensured. The robot can automatically adjust its operation mode and intensity according to the needs of the care recipient, reducing discomfort during the nursing process and improving the comfort of nursing care.

[0093] In this embodiment of the invention, the nursing transport mechanism consists of a second cylinder, a second regulating motor, a transport linkage unit, and a transport adsorption unit. Through the coordinated operation of the transport linkage unit and the transport adsorption unit, adaptive transport and adsorption of nursing items are achieved, meeting the collaborative nursing needs in different environments. This makes the transport process of nursing items more stable and efficient, greatly improving the efficiency and quality of nursing work, and providing strong support for intelligent companion robots and nursing collaboration systems.

[0094] In this embodiment of the invention, an Internet of Things (IoT) sensing module is provided. The IoT sensing module consists of a displacement sensor, a vision sensor, a voice sensor, and a pressure sensor, thereby covering multi-dimensional data such as the robot's motion state, the physiological and behavioral characteristics of the patient being cared for, environmental interaction information, and contact feedback. Through real-time monitoring and data stream transmission of the sensors, the IoT sensing module can dynamically capture changes in the nursing environment and, combined with real-time processing by the preprocessing unit, provide dynamic decision-making basis for the robot. This solves the pain point of poor environmental adaptability of traditional nursing equipment. By continuously monitoring key data, the preprocessing unit can mark abnormal data and transmit it to the decision control module, triggering early warnings or proactive nursing operations. This achieves a leap from passively responding to needs to proactively preventing risks, enhancing the predictability and safety of nursing services.

[0095] In this embodiment of the invention, the decision control module comprises a model training unit, a recognition and analysis unit, and a path planning unit. The path planning unit, combined with a multi-objective reward function, optimizes the path through proximal policy optimization (PPO). This multi-objective optimization mechanism considers not only positive rewards such as proximity to the target position, path smoothness, and task priority, but also negative penalties such as collision risk and deviation from the target, ensuring the optimality of the path. The collaborative cooperation of the model training unit, the recognition and analysis unit, and the path planning unit enables the robot to accurately identify the needs of the patient from the perceived data and generate corresponding nursing strategies and path plans. This collaborative cooperation ensures that the robot can make accurate decisions and flexibly execute nursing operations.

[0096] In this embodiment of the invention, when analyzing real-time datasets based on an intent recognition model, an autoencoder encodes the filtered multimodal data separately to accurately extract the latent feature vectors of each modality. A convolutional neural network (CNN) then performs convolutional fusion processing on the cross-modal aligned feature vector set. The sliding window operation of the convolutional kernel further extracts the fused modal features, and a gating fusion mechanism dynamically adjusts the weights of each modal feature. Combined with a Hidden Markov Model (HMM), the weighted modal features are fused and weighted, solving the problems of fixed weight allocation in traditional methods that cannot adapt to dynamic scenarios and ignore temporal dependencies between modalities. This improves the reliability and dynamic adaptability of the features. Finally, based on the Softmax activation function, a comprehensive intent probability distribution of the multimodal fused weighted features is output, providing a comprehensive intent recognition result. This probability distribution not only provides the most likely intent but also the probability of other possible intents, offering more reference information for subsequent path planning and action execution. Attached Figure Description

[0097] Figure 1 This is a schematic diagram of the structure of the intelligent companion robot and nursing collaboration system based on AI and IoT provided by the present invention.

[0098] Figure 2 This is a schematic diagram of the structure of the robot body provided by the present invention.

[0099] Figure 3 This is a three-dimensional structural diagram of the robot body provided by the present invention.

[0100] Figure 4 This is a front view of the robot body provided by the present invention.

[0101] Figure 5 This is a side view of the robot body provided by the present invention.

[0102] Figure 6 This is a schematic diagram of the structure of the nursing assistance arm provided by the present invention.

[0103] Figure 7 This is an isometric view of the nursing assistance arm provided by the present invention.

[0104] Figure 8 This is a side view of the nursing assistance arm provided by the present invention.

[0105] Figure 9 This is a schematic diagram of the adaptive clamping part provided by the present invention.

[0106] Figure 10 This is a schematic diagram of the nursing transport mechanism provided by the present invention.

[0107] Figure 11 This is an isometric drawing of the nursing transport mechanism provided by the present invention.

[0108] Figure 12 This is a side view of the nursing transport mechanism provided by the present invention.

[0109] Figure 13 This is a bottom view of the nursing transport mechanism provided by the present invention.

[0110] Figure 14 A schematic diagram illustrating the implementation process of a method for identifying and analyzing real-time datasets based on an intent recognition model is shown.

[0111] Figure 15 A schematic diagram illustrating the implementation process of a dynamic planning path generation method based on dynamic window path planning combined with intent recognition results is shown.

[0112] Figure 16 A schematic diagram illustrating the implementation process of an AI- and IoT-based intelligent companion robot and nursing collaboration method is shown.

[0113] In the diagram: 1-Robot body, 11-Main body base, 12-Angle adjustment motor, 13-Moving base, 14-Moving roller, 15-Lifting platform, 151-Vertical guide seat, 152-Vertical lifting seat, 16-Nursing workbench, 17-IoT sensing module, 171-Displacement sensor, 172-Vision sensor, 173-Voice sensor, 174-Pressure sensor, 175-Preprocessing unit, 2-Nursing assistance arm, 21-First swing motor, 22-First swing arm, 23-Second swing motor, 24-Second swing arm, 25-Third swing motor, 26-First cylinder, 27-Adaptive clamping part, 271-Clamping mounting base, 272-Clamping positioning slot, 273-First adjusting motor, 274-Adjusting screw, 275-Adaptive clamping 3-Nursing transfer mechanism, 31-Second cylinder, 32-T-shaped seat, 33-Second adjusting motor, 34-Transfer linkage part, 341-First hinge arm, 342-Third swing arm, 343-Lower hinge groove, 344-Second hinge arm, 345-Vertical seat, 346-Swing hinge sleeve, 347-First slide rod, 348-Hinge sleeve connecting seat, 349-Vibration damping protection spring, 35-Transfer moving part, 351-Second slide rod, 352-Connecting slide sleeve, 353-Third slide rod, 354-Transfer moving seat, 36-Transfer adsorption part, 361-Transfer adsorption seat, 362-Adsorption negative pressure groove, 400-Decision control module, 410-Model training unit, 420-Identification and analysis unit, 430-Path planning unit, 500-Action execution module. Detailed Implementation

[0114] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0115] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0116] Existing systems rely on a control module to determine different control modes for the nursing robot based on input information. The decision-making logic depends on preset rules rather than actively understanding the elderly person's intentions, resulting in weak intention understanding capabilities and insufficient adaptability, intelligence, and accuracy in human-machine collaboration in complex nursing scenarios. To address these issues, we propose an AI- and IoT-based intelligent companion robot and nursing collaboration system and method. In short, the system consists of an IoT sensing module 17, a decision control module 400, an action execution module 500, and an intelligent nursing robot. During operation, the IoT sensing module 17 first acquires real-time monitoring sensor data of the nursing environment and preprocesses it to obtain a real-time dataset. Then, the decision control module 400 acquires the real-time dataset, analyzes and identifies it based on an intention recognition model, generates an intention recognition result, and generates a dynamic planning path based on the dynamic window path planning method combined with the intention recognition result. The dynamic planning path is then used as an index to trigger nursing collaboration strategies in real time. Finally, the action execution module 500 acquires the nursing collaboration strategy, parses and identifies it, generates action execution instructions, and ultimately, the intelligent nursing robot responds to the action execution instructions to perform nursing collaboration work. In this embodiment of the invention, the IoT sensing module 17 collects multi-dimensional data such as the movements, facial expressions, voice commands, and contact pressure of the person being cared for in real time. Then, the AI ​​intent recognition model in the decision control module 400 performs in-depth analysis, thereby accurately inferring the elderly person's true care needs. Based on the intent recognition results, corresponding care strategies can be automatically generated without human intervention, improving the adaptability, safety, and service accuracy of the intelligent companion robot in the health and wellness scenario.

[0117] This invention provides an intelligent companion robot and nursing collaboration system based on AI and the Internet of Things. Figure 1 A schematic diagram of an AI- and IoT-based intelligent companion robot and nursing collaboration system is shown. The AI- and IoT-based intelligent companion robot and nursing collaboration system includes:

[0118] The Internet of Things sensing module 17 is used to acquire nursing environment monitoring sensor data, preprocess the monitoring sensor data, and obtain a real-time dataset.

[0119] The IoT sensing module 17 includes:

[0120] A displacement sensor 171, installed within the main body base 11, is used to sense the robot's displacement changes in real time, converting the displacement data into electrical signals and transmitting them. The displacement sensor 171 can sense the robot's displacement changes (such as movement speed, direction, and path trajectory) in real time and convert the displacement data into electrical signals for transmission. This data can be used to accurately track the robot's position in the care environment, ensuring it accurately reaches the target point (such as the bedside of an elderly person or the dressing bed area), and simultaneously provide real-time position feedback for dynamic window path planning, assisting in adjusting path segments to avoid collisions or deviations from the target.

[0121] A visual sensor 172 is used to acquire image information of the person being cared for, to identify the person's posture, facial expressions, skin condition, and the position and state of objects in the surrounding environment, and to generate visual monitoring data. Specifically, the visual sensor 172 acquires image information of the person being cared for through a camera, identifying their posture (e.g., raising a hand for help, turning over), facial expressions (e.g., pain, calmness), skin condition (e.g., pressure sores, redness and swelling), and the position and state of objects in the surrounding environment (e.g., the placement of dressing trays, the distribution of obstacles). For example, facial expression analysis can determine whether an elderly person is in pain or discomfort, and skin condition monitoring can identify pressure sore risk areas early.

[0122] A voice sensor 173, installed within the nursing workstation 16, is used to collect the voice commands and voice interaction information of the person being cared for in real time. Specifically, the voice sensor 173 can collect the voice commands (such as "help me wipe my back" or "call the nurse") and voice interaction information (such as a hurried tone indicating an urgent need, or a calm tone indicating a routine request). Through voice recognition and semantic understanding technology, it can quickly analyze the direct needs of the elderly, especially suitable for elderly people with limited mobility but clear verbal expression.

[0123] Pressure sensor 174 is used to collect the contact pressure between the sensor and the object being cared for in real time and convert the pressure data into an electrical signal for transmission.

[0124] The preprocessing unit 175 is communicatively connected to the displacement sensor 171, the vision sensor 172, the voice sensor 173, and the pressure sensor 174. It is used to receive displacement data, visual monitoring data, voice commands and voice interaction information, and pressure data, and integrate them into monitoring sensor data. It performs normalization and data cleaning preprocessing on the monitoring sensor data and outputs a real-time dataset.

[0125] In this embodiment of the invention, an Internet of Things (IoT) sensing module 17 is provided. The IoT sensing module 17 consists of a displacement sensor 171, a vision sensor 172, a voice sensor 173, and a pressure sensor 174, thereby covering multi-dimensional data such as robot motion status, physiological and behavioral characteristics of the patient being cared for, environmental interaction information, and contact feedback. Through real-time monitoring and data stream transmission of the sensors, the IoT sensing module 17 can dynamically capture changes in the nursing environment. Combined with the real-time processing of the preprocessing unit 175, it provides dynamic decision-making basis for the robot, solving the pain point of poor environmental adaptability of traditional nursing equipment. By continuously monitoring key data, the preprocessing unit 175 can mark abnormal data and transmit it to the decision control module 400, triggering early warning or proactive nursing operations. This realizes the leap from passively responding to needs to proactively preventing risks, enhancing the predictability and safety of nursing services.

[0126] The decision control module 400 is used to pre-build an AI-based intent recognition model, iteratively train the intent recognition model using historical datasets, output a converged intent recognition model, acquire real-time datasets, identify and analyze the real-time datasets based on the intent recognition model, generate intent recognition results, generate dynamic planning paths based on dynamic window path planning method combined with intent recognition results, and trigger nursing collaboration strategies in real time using dynamic planning paths as indexes.

[0127] The decision control module 400 includes:

[0128] Model training unit 410 is used to pre-build an AI-based intent recognition model, iteratively train the intent recognition model using historical datasets, and output a converged intent recognition model.

[0129] The intent recognition model is based on a convolutional neural network (CNN) model and includes an input layer and an output layer. A multimodal encoder is set between the input layer and the CNN model, and an adaptive filter is set between the multimodal encoder and the input layer. An attention mechanism is introduced in the multimodal encoder, which includes two Conv2D layers, a Transformer encoding layer, a fully connected layer, and an average pooling layer. A cross-modal attention mechanism is introduced in the CNN model. An LSTM layer is introduced between the CNN model and the output layer. The LSTM layer dynamically adjusts the weights of each modality feature based on a gating fusion mechanism and combines the Hidden Markov Model (HMM) algorithm to fuse and weight the modality features after weight adjustment to obtain multimodal fused weighted features. The intent comprehensive probability distribution of the multimodal fused weighted features is output based on the Softmax activation function.

[0130] The identification and analysis unit 420 identifies and analyzes the real-time dataset based on the intent recognition model and generates intent recognition results.

[0131] The path planning unit 430 generates a dynamic planning path based on the dynamic window path planning method combined with the intent recognition result, and triggers nursing collaboration strategies in real time using the dynamic planning path as an index.

[0132] In this embodiment of the invention, the decision control module 400 comprises a model training unit 410, a recognition and analysis unit 420, and a path planning unit 430. The path planning unit 430 optimizes the path using a proximal policy optimization (PPO) mechanism, incorporating a multi-objective reward function. This multi-objective optimization mechanism considers not only positive rewards such as proximity to the target position, path smoothness, and task priority, but also negative penalties such as collision risk and deviation from the target, ensuring the optimality of the path. The collaborative cooperation of the model training unit 410, the recognition and analysis unit 420, and the path planning unit 430 enables the robot to accurately identify the needs of the patient from the perceived data and generate corresponding nursing strategies and path plans. This collaborative cooperation ensures that the robot can make accurate decisions and flexibly execute nursing operations.

[0133] The action execution module 500 is used to acquire nursing collaboration strategies, parse and identify nursing collaboration strategies, and generate action execution instructions.

[0134] Intelligent nursing robots are used to respond to the action execution commands and perform nursing collaborative work.

[0135] In this embodiment, the system comprises an IoT sensing module 17, a decision control module 400, an action execution module 500, and an intelligent nursing robot. The IoT sensing module 17 acts as the system's "sensors," accurately capturing various monitoring and sensor data from the nursing environment. Through preprocessing, it transforms this data into a clear and orderly real-time dataset, providing a solid foundation for subsequent analysis and decision-making. The decision control module 400 is the system's "brain." It pre-builds an AI-based intent recognition model and iteratively trains it using historical datasets until a convergent and accurate intent recognition model is output. Subsequently, it acquires the real-time dataset, uses the model for deep recognition analysis, generates intent recognition results, and combines this with dynamic window path planning to plan the optimal dynamic planning path, thereby triggering corresponding nursing collaboration strategies in real time. The action execution module 500, as the system's "executor," receives the nursing collaboration strategy, performs detailed analysis and recognition, and generates clear action execution instructions. Finally, the intelligent nursing robot, as the "action taker" of the system, quickly responds to action execution commands, efficiently performs nursing collaboration work, and provides thoughtful and professional services to the nursing recipients. The IoT sensing module 17, decision control module 400, action execution module 500 and intelligent nursing robot are connected via Bluetooth or DTU communication.

[0136] In this embodiment of the invention, the IoT sensing module 17 collects multi-dimensional data such as the movements, facial expressions, voice commands, and contact pressure of the person being cared for in real time. Then, the AI ​​intent recognition model in the decision control module 400 performs in-depth analysis, thereby accurately inferring the elderly person's true care needs. Based on the intent recognition results, corresponding care strategies can be automatically generated without human intervention, improving the adaptability, safety, and service accuracy of the intelligent companion robot in the health and wellness scenario.

[0137] In a further preferred embodiment of the present invention, such as Figures 2-5 As shown, the intelligent nursing robot includes:

[0138] The robot body 1 includes a main body base 11, a lifting platform 15, and a nursing workbench 16. The lifting platform 15 is mounted on the main body base 11, and the top of the lifting platform 15 is detachably connected to the nursing workbench 16.

[0139] The robot body 1 also includes an angle adjustment motor 12 and a movable base 13. The angle adjustment motor 12 is fixedly installed at the bottom of the main body base 11, and the output end of the angle adjustment motor 12 is fixedly connected to the movable base 13. At least one set of movable rollers 14 are movably installed in the movable base 13, thereby facilitating the turning and movement of the robot body 1. The angle adjustment motor 12 can be a servo motor or a stepper motor. The lifting platform 15 consists of a lifting electric push rod, a vertical lifting seat 152, and a vertical guide seat 151. The vertical guide seat 151 is fixedly installed on the upper surface of the main body base 11. The lifting electric push rod is fixedly installed in the vertical guide seat 151. The telescopic end of the lifting electric push rod is fixedly connected to the vertical lifting seat 152. The vertical lifting seat 152 and the vertical guide seat 151 are slidably connected. The vertical lifting seat 152 is fixedly connected to the nursing workbench 16 by snap-fit ​​or riveting. The nursing workbench 16 can be a hollow rectangular seat or a "T" shaped seat structure.

[0140] At least one set of adjustable nursing assistance arms 2, said nursing assistance arms 2 are mounted on both sides of the nursing workbench 16;

[0141] The nursing transfer mechanism 3 works in conjunction with the nursing assistance arm 2. The nursing transfer mechanism 3 is located inside the nursing workbench 16 and is used for the automatic transfer and delivery of nursing supplies.

[0142] In this embodiment of the invention, the intelligent nursing robot consists of a robot body 1, a nursing assistance arm 2, and a nursing transfer mechanism 3. The nursing assistance arm 2 and the nursing transfer mechanism 3 work together to meet the requirements of companionship and nursing collaboration in different scenarios. The nursing transfer mechanism 3 is responsible for automatically transferring nursing supplies from the storage area to a designated location on the nursing workbench 16. The nursing assistance arm 2 accurately delivers items to the elderly person or directly assists them in using them, avoiding the tediousness of frequent manual retrieval and solving the problems of slow item retrieval and inaccurate positioning in traditional nursing care. This improves the response speed of basic nursing care. Furthermore, the collaborative work between the nursing assistance arm 2 and the nursing transfer mechanism 3 allows the robot to adapt to various nursing scenarios and meet the needs of different care recipients. The robot can automatically adjust its operation mode and item delivery method according to the specific scenario, ensuring the efficiency and accuracy of nursing operations and improving the adaptability and flexibility of nursing care. Through multimodal perception and adaptive adjustment mechanisms, the comfort and stability of nursing operations are ensured. The robot can automatically adjust its operation mode and intensity according to the needs of the care recipient, reducing discomfort during the nursing process and improving the comfort of nursing care.

[0143] In a further preferred embodiment of the present invention, such as Figures 6-8 As shown, the nursing assistance arm 2 includes:

[0144] The first swing motor 21 is fixedly installed inside the nursing workbench 16 by welding or riveting.

[0145] The first swing arm 22 is fixedly connected to the output shaft of the first swing motor 21;

[0146] The second swing motor 23 is fixedly installed at the end of the first swing arm 22, and the output shaft of the second swing motor 23 is fixedly connected to the second swing arm 24.

[0147] A third swing motor 25 is fixedly installed at the end of the second swing arm 24, and the output shaft of the third swing motor 25 is fixedly connected to the first cylinder 26.

[0148] The adaptive clamping part 27, fixedly connected to the telescopic rod of the first cylinder 26, is used to adaptively clamp and release care items. The adaptive clamping seat adjusts the operating parameters of the clamping seat motor in real time through a bidirectional coupled anti-resistance control algorithm. First, based on monitoring sensor data, and considering the combination of manual and visual methods, a receiver receiving intent coefficient is defined. , ,in express The cumulative duration of the caregiver's gaze at the clamping seat is displayed in the sliding window. Indicates the fixation duration threshold. Indicates the speed at which the person being cared for approaches with their hand. This indicates the maximum permissible hand approach speed. The adjustment coefficient is then used. Next, a weighted integral is performed on the rate of change in the gripping force of the person being cared for to quantify the smoothness of the interaction between the person being cared for and the gripper, defining the contact disturbance coefficient. This characterizes the smoothness of the interaction between the person being cared for and the clamp seat. ,in This indicates the rate of change of contact force of the clamping seat pressure sensor. Indicates the time forgetting factor, Represents the integral variable. The stability coefficient is used to identify smooth contact force as an effective contact force signal, and high-frequency oscillating force as an indicator that the person being cared for has not effectively grasped the item, thus serving as an invalid contact force signal. Next, the control stiffness of the gripper motor is defined using the receiver's intention coefficient and the contact disturbance coefficient. Stiffness indicates the clamping force of the clamping seat motor. ,in, Indicates the clamping stability reference stiffness. This indicates that the person being cared for is exerting an outward pulling force. Indicates standard interaction force, This indicates the minimum contact stiffness at which the clamping mechanism is held. This represents the activation function. This represents the disturbance sensitivity factor. Finally, based on the position increment integral structure and the contact force deviation and motor control stiffness, the position is updated according to the force state. The position update formula is: ,in Indicates the current position of the motor. Indicates the motor position after the command is executed. This represents the residual stress under interactive conditions. Indicates the position response parameters. This represents the integral variable, and the position is controlled by the motor to achieve a smooth interaction between the clamp and the person being cared for.

[0149] In this embodiment, the first swing motor 21, the second swing motor 23, the third swing motor 25, and the first cylinder 26 are communicatively connected to the action execution module 500. The action execution module 500 is used to control the opening, closing, and operation of the first swing motor 21, the second swing motor 23, the third swing motor 25, and the first cylinder 26. The first swing arm 22 and the second swing arm 24 can be rectangular arms or "T"-shaped arms, while the first cylinder 26 is fixedly installed at the end of the output shaft of the third swing motor 25 by a snap-fit ​​method.

[0150] In a further preferred embodiment of the present invention, such as Figure 9As shown, the adaptive clamping part 27 includes:

[0151] The clamping mounting base 271 is detachably mounted on the end of the telescopic rod of the first cylinder 26. The clamping mounting base 271 has at least one set of clamping positioning grooves 272. The clamping mounting base 271 can be a hollow rectangular or round structure. The clamping mounting base 271 is mounted on the end of the telescopic rod of the first cylinder 26 by snap-fit ​​or plug-in.

[0152] The first adjusting motor 273 is fixedly installed at the end of the clamping and positioning groove 272;

[0153] An adjusting screw 274 is rotatably installed in the clamping and positioning groove 272. One end of the adjusting screw 274 is fixedly connected to the adjusting motor. The number of clamping and positioning grooves 272 is 3-8 sets, and the clamping and positioning grooves 272 are evenly arranged axially. The adjusting screw 274 is rotatably connected to the clamping and positioning groove 272 by bearings or rollers.

[0154] An adaptive clamping seat 275 is threaded onto the outside of the adjusting screw 274, and the adaptive clamping seat 275 is slidably connected to the clamping positioning groove 272.

[0155] In this embodiment of the invention, an adaptive clamping part 27 is provided, which consists of a clamping mounting base 271, a first adjusting motor 273, an adjusting screw 274, and an adaptive clamping seat 275. Through the coordinated cooperation of each component, an adaptive clamping action is achieved for the object to be clamped. When the first adjusting motor 273 is started, the adjusting motor drives the adjusting screw 274 to rotate. Since the adaptive clamping seat 275 is threaded onto the outside of the adjusting screw 274 and slidably connected to the clamping positioning groove 272, the adaptive clamping seat 275 will move linearly along the clamping positioning groove 272 under the drive of the rotation of the adjusting screw 274. The distance between the two adaptive clamping seats 275 is adaptively adjusted according to the size of the object to be clamped, thereby completing the stable clamping of objects of different sizes and improving the applicability and flexibility of the nursing collaboration arm.

[0156] In a further preferred embodiment of the present invention, such as Figures 10-13 As shown, the nursing transport mechanism 3 includes:

[0157] The second cylinder 31 is fixedly installed on the surface of the nursing workbench 16. The second cylinder 31 is fixedly installed on the surface of the nursing workbench 16 by means of clamps or plug-in connection. The second cylinder 31 is connected to the action execution module 500.

[0158] A T-shaped seat 32 is fixedly connected to the telescopic rod of the second cylinder 31. The T-shaped seat 32 is slidably connected to the nursing workbench 16. A horizontal sliding groove for guiding and limiting the T-shaped seat 32 can be provided in the nursing workbench 16.

[0159] The second adjusting motor 33 is fixedly installed on the side wall of the T-shaped seat 32 by means of clamps or welding.

[0160] The transfer linkage 34 connected to the second regulating motor 33, and

[0161] The transfer adsorption unit 36 ​​is connected to the transfer linkage unit 34 and is used to adsorb care items.

[0162] In this embodiment of the invention, the nursing transport mechanism 3 consists of a second cylinder 31, a second regulating motor 33, a transport linkage unit 34, and a transport adsorption unit 36. Through the coordinated operation of the transport linkage unit 34 and the transport adsorption unit 36, adaptive transport and adsorption of nursing items are achieved, meeting the collaborative nursing needs in different environments. This makes the transport process of nursing items more stable and efficient, greatly improving the efficiency and quality of nursing work, and providing strong support for intelligent companion robots and nursing collaboration systems.

[0163] In a further preferred embodiment of the present invention, such as Figures 10-12 As shown, the transfer linkage 34 includes:

[0164] A first hinge arm 341 is fixedly connected to the output shaft of the second regulating motor 33, and a third swing arm 342 is provided on the side of the first hinge arm 341 away from the second regulating motor 33. The first hinge arm 341 is tenon-jointed or interference-fitted to the output shaft of the second regulating motor 33, and the end of the first hinge arm 341 away from the second regulating motor 33 is hinged to the lower hinge groove 343 of the third swing arm 342 through a hinge block.

[0165] The third swing arm 342 is rotatably mounted on the T-shaped seat 32, and the third swing arm 342 has an upper hinge groove and a lower hinge groove 343. One end of the first hinge arm 341 is hinged to the lower hinge groove 343, and the bottom of the third swing arm 342 is rotatably connected to the bottom of the T-shaped seat 32 through a bearing or roller.

[0166] The second hinge arm 344 is located on one side of the third swing arm 342, and one end of the second hinge arm 344 is hinged to the upper hinge groove. The other end of the second hinge arm 344 passes through the vertical stand 345 and is fixedly connected to the swing hinge sleeve 346.

[0167] The first slide rod 347 is slidably installed inside the swing hinge sleeve 346, and one end of the first slide rod 347 is fixedly connected to the hinge sleeve connecting seat 348.

[0168] A vibration damping and protection spring 349 is sleeved on the outside of the first slide rod 347, and one end of the vibration damping and protection spring 349 is fixedly connected to the swing hinge sleeve 346.

[0169] The transfer moving part 35 is detachably connected to the hinge sleeve connecting seat 348. The transfer moving part 35 includes a transfer moving seat 354, a second slide rod 351, a third slide rod 353, and a connecting slide sleeve 352. The transfer moving seat 354 is slidably sleeved on the second slide rod 351. The side wall of the transfer moving seat 354 is rotatably connected to the hinge sleeve connecting seat 348. One end of the rotating moving seat is detachably connected to a transfer adsorption part 36. The end of the second slide rod 351 is fixedly connected to the connecting slide sleeve 352. The connecting slide sleeve 352 is slidably sleeved on the third slide rod 353. The third slide rod 353 is detachably installed in the T-shaped seat 32.

[0170] The transport and adsorption section 36 includes:

[0171] The transfer adsorption seat 361 is detachably installed at the end of the transfer moving seat 354. The transfer adsorption seat 361 can be a hollow rectangular seat or a round seat, while the adsorption negative pressure groove 362 can be a round groove or a square groove. The negative pressure pump is communicatively connected to the action execution module 500.

[0172] At least one set of adsorption negative pressure tanks 362 are provided, the adsorption negative pressure tanks 362 are opened in the transfer adsorption seat 361, and a negative pressure pump is provided in the transfer adsorption seat 361, the negative pressure pump is connected to the adsorption negative pressure tanks 362.

[0173] In this embodiment of the invention, when adaptive transfer of nursing items is required, the action execution module 500 analyzes and identifies the nursing collaboration strategy, generates an action execution command, and activates the second cylinder 31. The second cylinder 31 moves the T-shaped seat 32, the transfer linkage part 34, and the transfer adsorption part 36 to the nursing item. Then, the second adjusting motor 33 is activated. The activation of the second adjusting motor 33 drives the first hinge arm 341 to rotate. The first hinge arm 341 drives the third swing arm 342 to swing. The swing of the third swing arm 342 drives the second hinge arm 344 to rotate. Arm 344 drives the swing hinge sleeve 346 to rotate. The swing hinge sleeve 346 drives the first slide rod 347 and the hinge sleeve connecting seat 348 to slide along the swing hinge sleeve 346, thereby causing the hinge sleeve connecting seat 348 to drive the transfer moving part 35 and the transfer adsorption part 36 to move. Then, the negative pressure pump is started. The negative pressure pump ensures that the adsorption negative pressure tank 362 forms a negative pressure environment, thereby realizing the adaptive picking and placing of nursing items. Then, through the second cylinder 31, the second adjusting motor 33, the swing of the third swing arm 342, and the coordinated cooperation of the second hinge arm 344, the nursing items are transferred to the target point.

[0174] This invention provides a method for identifying and analyzing real-time datasets based on an intent recognition model. Figure 14 The diagram illustrates the implementation flow of a method for identifying and analyzing real-time datasets based on an intent recognition model. This method includes:

[0175] S101: Obtain the real-time dataset, identify the data types in the real-time dataset, and perform adaptive filtering on the real-time data based on the data types to output the filtered dataset.

[0176] In adaptive filtering, the filtering formula is expressed as follows:

[0177]

[0178]

[0179] in, These are the filtered dataset and the original signal in the real-time dataset, respectively. These are the adaptive filter coefficients. For adaptive filtering time step, These are the filter coefficients from the previous time step. These are the step size parameter and the error signal, respectively.

[0180] S102: Load the filtered dataset, extract multimodal feature vectors from the filtered dataset based on the autoencoder, and perform cross-modal alignment of the feature vectors based on the attention mechanism, outputting a cross-modal aligned feature vector set. The self-attention mechanism (such as multi-head attention in Transformer) captures the global impact of the target point on path planning by calculating the correlation weights between the target point and all local obstacle points (such as medical staff, mobile devices, and furniture). For example, when the robot needs to bypass multiple obstacles to reach the target point, the self-attention mechanism can automatically identify the optimal detour order (such as prioritizing avoiding fast-moving medical staff, and then adjusting the path to avoid stationary furniture), avoiding the local shortsightedness problem of "only focusing on adjacent obstacles" in traditional methods.

[0181] S103, Obtain the feature vector set. The convolutional neural network model performs convolutional fusion processing on the feature vector set and extracts the fused modal features.

[0182] The fused modal features are represented as follows:

[0183]

[0184]

[0185] in, These are the fused modal features and the first convolutional result of the convolutional neural network model, respectively. This is the input representation of the feature vector set. These are the weight matrices of the first and second convolutional kernels in the convolutional neural network model, respectively.

[0186] S104: The weights of each modal feature are dynamically adjusted based on the gating fusion mechanism, and the modal features after weight adjustment are fused and weighted by the Hidden Markov algorithm to obtain multimodal fusion weighted features.

[0187] The multimodal fusion weighted features are represented as follows:

[0188]

[0189] in, These are visual modal features, speech modal features, and sensory modal features, respectively. These are the weights for visual modal features, speech modal features, and sensor modal features, respectively.

[0190]

[0191]

[0192]

[0193] The doorway mechanism score is based on visual modality features. For the gate mechanism score of speech modality features, The gate mechanism score is assigned to the sensor modal characteristics.

[0194] S105, based on the Softmax activation function, outputs the comprehensive probability distribution of intent based on the multimodal fusion weighted features, and outputs the intent recognition result, which includes the target location and operation type.

[0195] In this embodiment of the invention, when analyzing real-time datasets based on an intent recognition model, an autoencoder encodes the filtered multimodal data separately to accurately extract the latent feature vectors of each modality. A convolutional neural network (CNN) performs convolutional fusion processing on the cross-modal aligned feature vector set. The sliding window operation of the convolutional kernel further extracts the fused modal features, and a gating fusion mechanism dynamically adjusts the weights of each modal feature. Combined with a Hidden Markov Model (HMM), the weighted modal features are fused and weighted, solving the problems of fixed weight allocation in traditional methods being unable to adapt to dynamic scenarios and ignoring temporal dependencies between modalities. This improves the reliability and dynamic adaptability of the features. Finally, based on the Softmax activation function, the comprehensive intent probability distribution of the multimodal fused weighted features is output, providing a comprehensive intent recognition result. This probability distribution not only gives the most likely intent but also the probability of other possible intents, providing more reference information for subsequent path planning and action execution. A task dynamic scheduling algorithm is proposed. Based on actual nursing scenarios, a time- and space-based demand potential field is constructed to calculate task priorities. First, based on data from voice sensors, pressure sensors, and vision sensors, the first... Comprehensive nursing entropy of each monitoring node , ,in, , , These represent the speech weighting coefficient, the pressure weighting coefficient, and the image weighting coefficient, respectively. Indicates the first Data from a voice sensor, This represents the mean of the background noise. Indicates the first Data from a pressure sensor, Indicates the minimum pressure threshold. This represents the activation function. Indicates the first Confidence level of abnormal behavior identified by a visual sensor The standard deviation of environmental noise is represented by the comprehensive nursing entropy. The degree of current environmental anomaly is quantified using this comprehensive nursing entropy. Then, based on this comprehensive nursing entropy, distance attenuation and time attenuation are introduced to construct a demand potential energy field, and the comprehensive nursing potential energy at the anomaly source is calculated. ,in, Indicating intelligent companion robots Time for the first Comprehensive nursing potential of each abnormal source Indicates the time decay coefficient. Indicates the first The trigger time of each abnormal source Indicates the first The distance between the anomaly source and the location of the intelligent companion robot Indicates the spatial decay index, Let the spatial error coefficient be represented. Then, the task complexity coefficient is defined. The task cost is calculated through task consumption and operational risk, and a cost function is constructed. ,in, This indicates that intelligent nursing robots have reached the [number]th [number]. The overall execution cost of a task based on anomaly sources. This indicates that intelligent nursing robots have reached the [number]th [number]. Calculate the path planning distance for the task based on the anomaly source. This indicates the battery percentage of the intelligent nursing robot. Indicates the first The task complexity coefficient of each abnormal source task , , This represents the cost weighting coefficient. Finally, based on the comprehensive nursing potential and comprehensive execution cost ratio, a nonlinear gain mechanism is used to amplify the task advantages, improve the sensitivity of nursing tasks, and construct a global evaluation index for nursing tasks. , ,in, This represents the total number of tasks, based on the overall evaluation index of nursing tasks. This reflects the importance of intelligent nursing robots for the first The probability of task activation for each abnormal source.

[0196] This invention provides a method for generating dynamic planning paths based on dynamic window path planning combined with intent recognition results. Figure 15 The diagram illustrates the implementation flow of a method for generating dynamic planning paths based on dynamic window path planning combined with intent recognition results. This method includes:

[0197] S201, Obtain the intent recognition result, determine the target point spatial coordinates, operation type, type priority, and target point real-time environmental data based on the intent recognition result, map the target point spatial coordinates, operation type, type priority, and target point real-time environmental data to the state space, and convert the data in the state space into a high-dimensional feature vector through the decoder;

[0198] S202 loads high-dimensional feature vectors, captures the global dependency between the target point and local obstacle points through a self-attention mechanism, and gradually generates the initial path between the target point and the robot's current coordinates through a generative adversarial network.

[0199] S203, obtain the initial path between the current coordinate point and the target point, and adjust the path segments of the initial path by combining the global dependency relationship between the target point and local obstacle points through dynamic window path planning method to obtain the adjusted path;

[0200] S204 employs a multi-objective reward function, which uses proximity to the target location, path smoothness, and priority as positive rewards, and collision risk and deviation from the target as negative penalties. This function is constructed by adjusting the path's waypoints. Proximal Policy Optimization (PPO) is used to solve this multi-objective reward function, generating a dynamic programming path. PPO iteratively updates the robot's path selection strategy, continuously trying different path combinations and evaluating reward feedback in a simulated environment, gradually converging to the optimal strategy. This balances exploration (trying new paths) with utilization (selecting known, relatively good paths), ensuring that a globally optimal or near-optimal dynamic programming path can still be found in complex nursing scenarios (such as multiple obstacles or dynamic environments).

[0201] The multi-objective reward function is expressed as:

[0202]

[0203] in, These are, respectively, the target proximity weight coefficient, the path smoothness weight coefficient, the task priority weight coefficient, the collision risk weight coefficient, and the deviation from the target weight coefficient. Each represents the current time. Robot's current coordinates, target coordinates, and previous time point. Robot coordinates, next moment Robot coordinates The set of obstacle points at the current moment; These are reward items for target proximity, path smoothness, task priority, collision risk, and deviation from the target.

[0204] In this embodiment of the invention, when generating a dynamically planned path based on the dynamic window path planning method combined with the intent recognition result, the decoder converts the data in the state space into a high-dimensional feature vector, enabling the method to generate richer feature representations. This high-dimensional feature representation can capture complex environmental information and task requirements, providing more comprehensive data support for subsequent path generation and adjustment. Furthermore, by capturing the global dependency relationship between the target point and local obstacle points through a self-attention mechanism, the method can comprehensively consider the mutual influence between the target point and obstacles. By adjusting the path segments of the initial path based on the global dependency relationship between the target point and local obstacle points using the dynamic window path planning method, the method can dynamically adjust the path to adapt to environmental changes. The dynamic window path planning method can quickly generate and adjust paths within a local window, ensuring the real-time performance and adaptability of the path. Finally, based on Proximal Policy Optimization (PPO), the multi-objective reward function is solved, allowing the method to dynamically adjust the path planning strategy and ensure the optimality of the path planning. The PPO algorithm can gradually optimize the path planning strategy while maintaining policy stability, improving the efficiency and quality of path planning.

[0205] On the other hand, embodiments of the present invention also provide a method for intelligent companion robots and nursing collaboration based on AI and the Internet of Things. Figure 16 A schematic diagram illustrating the implementation process of an AI- and IoT-based intelligent companion robot and nursing collaboration method is shown. This AI- and IoT-based intelligent companion robot and nursing collaboration method includes:

[0206] S10: Acquire nursing environment monitoring sensor data, preprocess the monitoring sensor data to obtain real-time dataset;

[0207] S20: A pre-built AI-based intent recognition model is used to iteratively train the intent recognition model using historical datasets, and a converged intent recognition model is output.

[0208] S30: Obtain real-time dataset, identify and analyze the real-time dataset based on the intent recognition model, generate intent recognition results, generate dynamic planning path based on dynamic window path planning method combined with intent recognition results, and trigger nursing collaboration strategy in real time with dynamic planning path as index.

[0209] S40: Obtain nursing collaboration strategy, analyze and identify nursing collaboration strategy, generate action execution instructions, and the intelligent nursing robot responds to the action execution instructions to perform nursing collaboration work.

[0210] In summary, this invention provides an intelligent companion robot and nursing collaboration system and method based on AI and IoT. In the embodiments of this invention, the IoT sensing module 17 collects multi-dimensional data such as the movement posture, facial expression, voice commands, and contact pressure of the person being cared for in real time. Then, the AI ​​intent recognition model in the decision control module 400 performs in-depth analysis, thereby accurately inferring the elderly person's real nursing needs. The corresponding nursing strategy can be automatically generated based on the intent recognition results without human intervention, thus improving the adaptability, safety, and service accuracy of the intelligent companion robot in the health and wellness scenario.

[0211] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0212] It should be understood that the disclosed apparatus can be implemented in other ways, given the several embodiments provided in this application. For example, the apparatus embodiments described above are merely illustrative; the division of units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or communication connections shown or discussed may be through some interfaces; the indirect coupling or communication connections between devices or units may be telecommunications or other forms.

[0213] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0214] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. An intelligent companion robot and nursing collaboration system based on AI and IoT, characterized in that, include: The Internet of Things (IoT) sensing module is used to acquire monitoring sensor data of the nursing environment, preprocess the monitoring sensor data, and obtain a real-time dataset. The decision control module is used to pre-build an AI-based intent recognition model, iteratively train the intent recognition model using historical datasets, output a converged intent recognition model, acquire real-time datasets, identify and analyze the real-time datasets based on the intent recognition model, generate intent recognition results, generate dynamic planning paths based on dynamic window path planning method combined with intent recognition results, and trigger nursing collaboration strategies in real time using dynamic planning paths as indexes. The action execution module is used to acquire nursing collaboration strategies, parse and identify the nursing collaboration strategies, and generate action execution instructions; An intelligent nursing robot is used to respond to the action execution command and perform collaborative nursing tasks; the intelligent nursing robot includes: The robot body includes a main body base, a lifting platform, and a nursing workbench. The lifting platform is mounted on the main body base, and its top is detachably connected to the nursing workbench. At least one set of adjustable nursing assistance arms are mounted on both sides of the nursing workbench. A nursing transfer mechanism, working in conjunction with the nursing assistance arms, is located within the nursing workbench and is used for the automatic transfer and delivery of nursing items. Each nursing assistance arm includes: a first swing motor, fixedly mounted within the nursing workbench; a first swing arm fixedly connected to the output shaft of the first swing motor; a second swing motor, fixedly mounted at the end of the first swing arm, with its output shaft fixedly connected to the second swing arm; a third swing motor fixedly mounted at the end of the second swing arm, with its output shaft fixedly connected to a first cylinder; and an adaptive clamping part fixedly connected to the telescopic rod of the first cylinder, used for adaptively clamping and releasing nursing items. The adaptive clamping part includes: A clamping mounting base is detachably mounted on the end of the first cylinder telescopic rod, and at least one set of clamping positioning grooves is provided in the clamping mounting base; a first adjusting motor is fixedly mounted on the end of the clamping positioning groove; an adjusting screw is rotatably mounted in the clamping positioning groove, and one end of the adjusting screw is fixedly connected to the adjusting motor; an adaptive clamping seat is threaded onto the outside of the adjusting screw, and the adaptive clamping seat is slidably connected to the clamping positioning groove; The IoT sensing module includes: The displacement sensor, installed inside the main body base, is used to sense the robot's displacement changes in real time, convert the displacement data into electrical signals and transmit them. Visual sensors are used to collect image information of the person being cared for, in order to identify the person's movements, facial expressions, skin condition, as well as the position and state of objects in the surrounding environment, and generate visual monitoring data. A voice sensor, installed inside the nursing workstation, is used to collect the voice commands and voice interaction information of the person being cared for in real time; Pressure sensors are used to collect the contact pressure between the object being cared for in real time and convert the pressure data into electrical signals for transmission. The preprocessing unit is communicatively connected to the displacement sensor, vision sensor, voice sensor, and pressure sensor. It is used to receive displacement data, visual monitoring data, voice commands and voice interaction information, and pressure data, and integrate them into monitoring sensor data. It performs normalization and data cleaning preprocessing on the monitoring sensor data and outputs a real-time dataset. The adaptive clamping holder uses a bidirectional coupled anti-resistance control algorithm to adjust the operating parameters of the clamping holder motor in real time. First, based on monitoring sensor data, and considering the combination of manual and visual methods, a receiver intention coefficient is defined. , ,in express The cumulative gaze duration of the person being cared for regarding the clamped area is displayed in the sliding window. Indicates the fixation duration threshold. Indicates the speed at which the person being cared for approaches with their hand. This indicates the maximum permissible hand approach speed. The adjustment coefficient is then expressed. Next, a weighted integral is performed on the rate of change in the gripping force of the person being cared for to quantify the smoothness of the interaction between the person being cared for and the gripping part, defining the contact disturbance coefficient. This characterizes the smoothness of the interaction between the person being cared for and the clamp seat. ,in This indicates the rate of change of contact force of the clamping seat pressure sensor. Indicates the time forgetting factor, Represents the integral variable. Indicates the current time of the sliding window. The stability coefficient is used to identify smooth contact force as an effective contact force signal, and high-frequency oscillating force as an indicator that the person being cared for has not effectively grasped the item, thus serving as an invalid contact force signal. Then, the control stiffness of the gripper motor is defined using the receiver's intention coefficient and the contact disturbance coefficient. Stiffness indicates the clamping force of the clamping seat motor. ,in, Indicates the clamping stability reference stiffness. This indicates that the person being cared for is exerting an outward pulling force. Indicates standard interaction force, This indicates the minimum contact stiffness at which the clamping mechanism is held. This represents the activation function. This represents the disturbance sensitivity factor. Finally, based on the position increment integral structure and the contact force deviation and motor control stiffness, the position is updated according to the force state. The position update formula is: ,in Indicates the current position of the motor. Indicates the motor position after the command is executed. This represents the residual stress under interactive conditions. Indicates the position response parameters. This represents the integral variable, and the position is controlled by the motor to achieve a smooth interaction between the clamp and the items being cared for. The decision control module includes: The model training unit is used to pre-build an AI-based intent recognition model, iteratively train the intent recognition model using historical datasets, and output a converged intent recognition model. The intent recognition model is based on a convolutional neural network (CNN) model and includes an input layer and an output layer. A multimodal encoder is set between the input layer and the CNN model, and an adaptive filter is set between the multimodal encoder and the input layer. An attention mechanism is introduced into the multimodal encoder, which includes two Conv2D layers, a Transformer encoding layer, a fully connected layer, and an average pooling layer. A cross-modal attention mechanism is introduced into the CNN model. An LSTM layer is introduced between the CNN model and the output layer. The LSTM layer dynamically adjusts the weights of each modality feature based on a gating fusion mechanism and combines the Hidden Markov Model (HMM) algorithm to fuse and weight the modality features after weight adjustment to obtain multimodal fused weighted features. The intent comprehensive probability distribution of the multimodal fused weighted features is output based on the Softmax activation function. A task dynamic scheduling algorithm is proposed to perform task priority planning. The identification and analysis unit identifies and analyzes real-time datasets based on the intent recognition model, and generates intent recognition results. The path planning unit generates a dynamic planning path based on the dynamic window path planning method combined with the intent recognition result, and uses the dynamic planning path as an index to trigger nursing collaboration strategies in real time.

2. The intelligent companion robot and nursing collaboration system based on AI and IoT as described in claim 1, characterized in that: The nursing transport facility includes: The second cylinder is fixedly installed on the surface of the nursing workbench; A T-shaped seat is fixedly connected to the telescopic rod of the second cylinder, and the T-shaped seat is slidably connected to the nursing workbench; The second adjusting motor is fixedly installed on the side wall of the T-shaped seat; The transfer linkage unit is connected to the second regulating motor, and the transfer adsorption unit is connected to the transfer linkage unit for adsorbing care items.

3. The intelligent companion robot and nursing collaboration system based on AI and IoT as described in claim 2, characterized in that: The transfer linkage unit includes: A first hinge arm is fixedly connected to the output shaft of a second regulating motor, and a third swing arm is provided on the side of the first hinge arm away from the second regulating motor. The third swing arm is rotatably mounted on the T-shaped seat, and the third swing arm has an upper hinge groove and a lower hinge groove. One end of the first hinge arm is hinged to the lower hinge groove. The second hinge arm is located on one side of the third swing arm, and one end of the second hinge arm is hinged to the upper hinge groove. The other end of the second hinge arm passes through the vertical stand and is fixedly connected to a swing hinge sleeve. The first slide rod is slidably installed inside the swing hinge sleeve, and one end of the first slide rod is fixedly connected to the hinge sleeve connecting seat; A vibration damping and protection spring is sleeved on the outside of the first slide rod, and one end of the vibration damping and protection spring is fixedly connected to the swing hinge sleeve; A transfer moving part is detachably connected to the hinged sleeve connecting seat. The transfer moving part includes a transfer moving seat, a second slide rod, a third slide rod, and a connecting slide sleeve. The transfer moving seat is slidably sleeved on the second slide rod. The side wall of the transfer moving seat is rotatably connected to the hinged sleeve connecting seat. One end of the rotatable moving seat is detachably connected to a transfer adsorption part. The end of the second slide rod is fixedly connected to a connecting slide sleeve. The connecting slide sleeve is slidably sleeved on the third slide rod. The third slide rod is detachably installed in the T-shaped seat. The transport and adsorption section includes: A transfer adsorption seat, which is detachably mounted on the end of a transfer movable seat; At least one set of adsorption negative pressure tanks are provided, the adsorption negative pressure tanks are opened in the transfer adsorption seat, and a negative pressure pump is provided in the transfer adsorption seat, the negative pressure pump is connected to the adsorption negative pressure tanks.

4. The intelligent companion robot and nursing collaboration system based on AI and IoT as described in claim 3, characterized in that: The method for identifying and analyzing real-time datasets based on an intent recognition model includes: Acquire a real-time dataset, identify the data types in the real-time dataset, perform adaptive filtering on the real-time data based on the data types, and output the filtered dataset. Load the filtered dataset, extract the modal feature vectors of the multimodal data from the filtered dataset based on the autoencoder, and perform cross-modal alignment of the modal feature vectors based on the attention mechanism, and output the cross-modal aligned feature vector set; The feature vector set is obtained, and the convolutional neural network model performs convolutional fusion processing on the feature vector set to extract the fused modal features. The weights of each modal feature are dynamically adjusted based on the gating fusion mechanism, and the weighted modal features after weight adjustment are fused and weighted by the Hidden Markov algorithm to obtain multimodal fusion weighted features. Based on the Softmax activation function, the comprehensive probability distribution of intent is output using multimodal fusion weighted features, and the intent recognition result is output, which includes target location and operation type. A dynamic task scheduling algorithm is proposed. Based on actual nursing scenarios, a time- and space-based demand potential field is constructed to calculate task priorities. First, based on data from voice, pressure, and vision sensors, the first... Comprehensive nursing entropy of each monitoring node , ,in, , , These represent the speech weighting coefficient, the pressure weighting coefficient, and the image weighting coefficient, respectively. Indicates the first Data from a voice sensor, This represents the mean of the background noise. Indicates the first Data from a pressure sensor, Indicates the minimum pressure threshold. This represents the activation function. Indicates the first Confidence level of abnormal behavior identified by a visual sensor The standard deviation of environmental noise is represented by the comprehensive nursing entropy. The degree of current environmental anomaly is quantified using this comprehensive nursing entropy. Then, based on this comprehensive nursing entropy, distance attenuation and time attenuation are introduced to construct a demand potential energy field, and the comprehensive nursing potential energy at the anomaly source is calculated. ,in, Indicating intelligent companion robots Time for the first Comprehensive nursing potential of each abnormal source Indicates the time decay coefficient. Indicates the first The trigger time of each abnormal source Indicates the first The distance between the anomaly source and the location of the intelligent companion robot Indicates the spatial decay index, Let the spatial error coefficient be represented. Then, the task complexity coefficient is defined. The task cost is calculated through task consumption and operational risk, and a cost function is constructed. ,in, This indicates that intelligent nursing robots have reached the [number]th [number]. The overall execution cost of a task based on anomaly sources. This indicates that intelligent nursing robots have reached the [number]th [number]. Calculate the path planning distance for the task based on the anomaly source. This indicates the battery percentage of the intelligent nursing robot. Indicates the first The task complexity coefficient of each abnormal source task , , This represents the cost weighting coefficient. Finally, based on the comprehensive nursing potential and comprehensive execution cost ratio, a nonlinear gain mechanism is used to amplify the task advantages, improve the sensitivity of nursing tasks, and construct a global evaluation index for nursing tasks. , ,in, This represents the total number of tasks, based on the overall evaluation index of nursing tasks. This reflects the importance of intelligent nursing robots for the first The probability of task activation for each abnormal source.

5. The intelligent companion robot and nursing collaboration system based on AI and IoT as described in claim 4, characterized in that: The method for generating dynamic planning paths based on dynamic window path planning combined with intent recognition results includes: The intent recognition result is obtained, and the target point spatial coordinates, operation type, type priority, and target point real-time environmental data are determined based on the intent recognition result. The target point spatial coordinates, operation type, type priority, and target point real-time environmental data are mapped to the state space, and the data in the state space is converted into a high-dimensional feature vector through the decoder. Load high-dimensional feature vectors, capture the global dependency between the target point and local obstacle points through a self-attention mechanism, and gradually generate the initial path between the target point and the robot's current coordinates through a generative adversarial network; Obtain the initial path between the current coordinate point and the target point. Then, adjust the path segments of the initial path by combining the global dependency relationship between the target point and local obstacle points using the dynamic window path planning method to obtain the adjusted path. The system uses proximity to the target location, path smoothness, and priority as positive rewards, and collision risk and deviation from the target as negative penalties. It constructs a multi-objective reward function by combining the waypoints of the adjusted path, and solves the multi-objective reward function based on the near-end strategy optimization to generate a dynamic programming path.

6. A method for intelligent companion robots and nursing collaboration based on AI and IoT, implemented using the intelligent companion robot and nursing collaboration system based on AI and IoT as described in any one of claims 1-5, characterized in that: The AI- and IoT-based intelligent companion robot and nursing collaboration method includes: Acquire nursing environment monitoring sensor data, preprocess the monitoring sensor data to obtain real-time dataset; A pre-built AI-based intent recognition model is used to iteratively train the intent recognition model using historical datasets, and a converged intent recognition model is output. Acquire real-time datasets, analyze and identify the real-time datasets based on intent recognition models, generate intent recognition results, generate dynamic planning paths based on dynamic window path planning methods combined with intent recognition results, and trigger nursing collaboration strategies in real time using dynamic planning paths as indexes. The system acquires nursing collaboration strategies, analyzes and identifies these strategies, generates action execution instructions, and the intelligent nursing robot responds to these instructions to perform nursing collaboration tasks.

Images