Nursing robot motion prediction control method and system based on cloud platform

Abstract

The invention discloses a nursing robot motion prediction control method and system based on a cloud platform, and relates to the technical field of robot nursing, and the method comprises the steps: obtaining the motion feature data of a nursing robot and a patient; decomposing the motion feature data of the nursing robot and the patient based on a singular value decomposition algorithm, and extracting dominant and invisible features to obtain a comprehensive motion feature data set; uploading the comprehensive motion feature data set to a cloud platform through an Internet of Things gateway, and training a motion prediction model based on the comprehensive motion feature data set in the cloud platform; outputting the next motion of the nursing robot according to the motion prediction model; and based on the output motion track and posture data, generating a corresponding control instruction to realize motion control of the nursing robot. According to the method, the dominant features and the recessive features are extracted, the data availability is improved, the multi-layer perceptron is used for motion prediction, and the prediction accuracy can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot nursing, in particular to a nursing robot motion prediction control method and system based on a cloud platform. BACKGROUND

[0002] A nursing robot is an intelligent robot designed to assist nursing staff and patients in completing various nursing tasks. It is used in hospitals, nursing homes, or family environments. The nursing robot can assist patients in daily activities such as moving, eating, bathing, and taking medicine, reducing the workload of nursing staff and improving the quality and efficiency of nursing services. By predicting the future motion state of the robot, its control strategy can be optimized. This method not only considers current sensor data and control instructions, but also predicts future state changes through mathematical models and algorithms, achieving more accurate and efficient motion control.

[0003] The sensor data and control algorithms of the robot are processed and stored in the cloud, taking advantage of the powerful computing power and massive data resources of cloud computing to perform more complex and accurate motion prediction and control. On the cloud platform, real-time data analysis, machine learning, and dynamic optimization can be performed to improve the intelligence and adaptability of the robot. In addition, cloud-based solutions can also enable multi-robot collaboration, remote monitoring and maintenance, further improving the efficiency and flexibility of nursing services.

[0004] In existing nursing robot control technology, for example, Chinese patent 201410834292.9 discloses a control method and device for a rehabilitation nursing robot bed, which first sets motion instructions and performs robot action execution according to the motion instructions, reducing medical staff and freeing up labor.

[0005] However, existing nursing robots still need to address the following issues:

[0006] (1) How to effectively obtain the motion feature data of the nursing robot and the patient.

[0007] (2) How to extract useful information from the nursing robot and the patient, including explicit motion features and hidden motion patterns.

[0008] (3) How to use the extracted feature data to predict the future motion of the nursing robot.

[0009] In view of the problems in the related art, no effective solutions have been proposed so far. SUMMARY

[0010] In view of the problems in the prior art, the present application provides a nursing robot motion prediction control method and system based on a cloud platform to overcome the above technical problems existing in the prior art.

[0011] To this end, the present application adopts the following specific technical solutions:

[0012] According to one aspect of the present application, a nursing robot motion prediction control method based on a cloud platform is provided, which comprises the following steps:

[0013] S1, obtain the motion characteristic data of the nursing robot and the patient, and the motion characteristic data of the nursing robot and the patient includes robot position data, robot speed data and patient posture data.

[0014] S2, decompose the motion characteristic data of the nursing robot and the patient based on a singular value decomposition algorithm, and extract explicit and implicit features to obtain a comprehensive motion characteristic data set.

[0015] S3, upload the comprehensive motion characteristic data set to the cloud platform through an Internet of Things gateway, and train a motion prediction model based on the comprehensive motion characteristic data set in the cloud platform.

[0016] S4, output the next step motion of the nursing robot according to the motion prediction model, and the next step motion of the nursing robot includes trajectory data and posture data.

[0017] S5, generate corresponding control instructions based on the output motion trajectory and posture data to realize motion control of the nursing robot.

[0018] Further, obtaining the motion characteristic data of the nursing robot and the patient comprises the following steps:

[0019] S11, obtain the robot position data, robot speed data and patient posture data, and the robot speed data includes linear speed and angular speed;

[0020] S12, data filtering, normalization and time alignment are performed on the robot position data, robot speed data and patient posture data.

[0021] Further, decomposing the motion characteristic data of the nursing robot and the patient based on a singular value decomposition algorithm and extracting explicit and implicit features to obtain a comprehensive motion characteristic data set comprises the following steps:

[0022] S21, construct the robot position data, speed data and patient posture data into a multi-dimensional data matrix, wherein the rows represent different time points and the columns represent different robot position data, robot speed data and patient posture data;

[0023] S22, performing singular value decomposition on the multi-dimensional data matrix by using a singular value decomposition algorithm to obtain left singular vectors, right singular vectors and singular values;

[0024] S23, extracting explicit features and implicit features from the decomposition result;

[0025] S24, re-integrating the explicit features and the implicit features into a comprehensive motion feature dataset.

[0026] Further, performing singular value decomposition on the multi-dimensional data matrix by using a singular value decomposition algorithm to obtain left singular vectors, right singular vectors and singular values includes the following steps:

[0027] S221, selecting a singular value decomposition algorithm and setting parameters for singular value decomposition, including the number of singular values to be decomposed;

[0028] S222, performing singular value decomposition on the multi-dimensional data matrix to obtain left singular vectors, singular values and right singular vectors;

[0029] The left singular vector is used to reveal the motion pattern of the nursing robot and the patient over time;

[0030] The singular value is used to measure the importance of each feature and the corresponding time pattern, and the greater the singular value, the greater the influence of the corresponding feature on the overall motion;

[0031] The right singular vector includes the contribution size of the robot position data, the robot speed data and the patient posture data, and is used to identify the influence degree of each feature in the motion pattern.

[0032] Further, extracting explicit features and implicit features from the decomposition result includes the following steps:

[0033] S231, identifying the principal components of the robot position data, the robot speed data and the patient posture data by analyzing the singular values and the right singular vectors to obtain explicit features;

[0034] S232, sorting the explicit features according to the size of the singular values and selecting a number of explicit features;

[0035] S233, extracting the patterns and trends of the robot and the patient hidden in the data by analyzing the singular values and the right singular vectors to obtain implicit features.

[0036] Further, extracting the patterns and trends hidden in the data by analyzing the singular values and the right singular vectors to obtain implicit features includes the following steps:

[0037] S2331, performing data dimensionality reduction on the right singular vector based on the size of the singular value;

[0038] S2332, cluster the right singular vectors using a clustering algorithm to identify different patterns and trends;

[0039] S2333, time series analysis of the left singular vectors to identify implicit feature patterns that change over time;

[0040] S2334, combine the clustering results in S2332 with the time series analysis in S2333 to generate implicit features.

[0041] Further, through the Internet of Things gateway, upload the comprehensive motion feature data set to the cloud platform, and in the cloud platform, train the motion prediction model based on the comprehensive motion feature data set, including the following steps:

[0042] S31, upload the encrypted comprehensive motion feature data set to the cloud platform through the Internet of Things gateway, decrypt the received comprehensive motion feature data set in the cloud platform, and store it in the database of the cloud platform;

[0043] S32, obtain the label data corresponding to the comprehensive motion feature data set, encode the label data, and select a multi-layer perceptron for motion prediction;

[0044] S33, train and evaluate the multi-layer perceptron using the comprehensive motion feature data set containing the label data to obtain the motion prediction model;

[0045] S34, package the trained motion prediction model and transmit it to the local system of the nursing robot;

[0046] Wherein, the loss function of the motion prediction model is:

[0047]

[0048] In the formula, x i represents the i-th motion data of the nursing robot predicted, y i represents the i-th motion data of the nursing robot actually occurred;

[0049] m represents the number of training samples.

[0050] Further, the way to obtain the label data corresponding to the comprehensive motion feature data set includes expert annotation, sensor data fusion, scene recording and user feedback.

[0051] Further, according to the motion prediction model, output the next step motion of the nursing robot, and the next step motion of the nursing robot includes trajectory data and posture data, including the following steps:

[0052] S41, acquire the latest comprehensive motion feature data set, and extract the comprehensive motion feature data at the current moment as input data;

[0053] S42, substitute the input data into the trained motion prediction model, and calculate the next step motion prediction result of the nursing robot, including trajectory data and posture data.

[0054] Further, based on the output motion trajectory and posture data, corresponding control instructions are generated to realize the motion control of the nursing robot, including the following steps:

[0055] S51, generate corresponding robot control instructions based on the predicted motion trajectory and posture data;

[0056] S52, combine the environmental data to plan the path, so as to ensure that the nursing robot avoids obstacles in the execution process;

[0057] S53, perform collision detection and redundancy check on the generated robot control instructions to ensure safety.

[0058] According to another aspect of the present application, a cloud platform-based nursing robot motion prediction control system is also provided, which comprises a data acquisition module, a data decomposition module, a model training module, a motion prediction module and a robot control module; the data acquisition module, the data decomposition module, the model training module, the motion prediction module and the robot control module are sequentially connected.

[0059] The data acquisition module is used to acquire the motion feature data of the nursing robot and the patient, and the motion feature data of the nursing robot and the patient comprises robot position data, robot speed data and patient posture data. The data decomposition module is used to decompose the motion feature data of the nursing robot and the patient based on a singular value decomposition algorithm, and extract explicit and implicit features to obtain a comprehensive motion feature data set. The model training module is used to upload the comprehensive motion feature data set to a cloud platform through an Internet of Things gateway, and train a motion prediction model in the cloud platform based on the comprehensive motion feature data set. The motion prediction module is used to output the next step motion of the nursing robot according to the motion prediction model, and the next step motion of the nursing robot comprises trajectory data and posture data. The robot control module is used to generate corresponding control instructions based on the output motion trajectory and posture data, and realize the motion control of the nursing robot.

[0060] The present application has the following advantages:

[0061] (1) The application collects multi-dimensional motion feature data of the robot and the patient, and performs dimension reduction processing on the multi-dimensional motion feature data through singular value decomposition, removes noise and redundant information, improves the quality and processing efficiency of the data. Extracting explicit and implicit features, identifying the main patterns and potential relationships in the data, providing a more representative and generalizable feature set for the motion prediction model. Data decomposition and feature extraction can compress the data volume without losing key information, reducing the burden of subsequent calculation and storage.

[0062] (2) Through the Internet of Things gateway, the data is transmitted efficiently, ensuring that the data can be uploaded to the cloud platform in time for processing. Using the powerful computing power and resources of the cloud platform, complex motion prediction model training is performed to improve the accuracy and stability of the model. The cloud platform can realize continuous data updating and model optimization, so that the motion prediction model can continuously learn and adapt to new data and changes.

[0063] (3) Based on the trained motion prediction model, the next step of the nursing robot can be accurately predicted, that is, the robot's motion is planned in advance to reduce the occurrence of unexpected situations. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor based on these drawings.

[0065] Figure 1 is a flow chart of a nursing robot motion prediction control method based on a cloud platform according to an embodiment of the present application;

[0066] Figure 2 is a principle block diagram of a nursing robot motion prediction control system based on a cloud platform according to an embodiment of the present application.

[0067] In the figure:

[0068] 1, data acquisition module; 2, data decomposition module; 3, model training module; 4, motion prediction module; 5, robot control module. DETAILED DESCRIPTION

[0069] For further illustrating the embodiments, the present application provides the accompanying drawings, which are part of the present application disclosure, mainly used to illustrate the embodiments and can be used to explain the operation principle of the embodiments in conjunction with the relevant description of the specification. Those skilled in the art can understand other possible implementations and advantages of the present application by referring to these contents. The components in the drawings are not drawn to scale, and similar component symbols are usually used to represent similar components.

[0070] According to an embodiment of the present application, a cloud platform-based nursing robot motion prediction control method and system are provided.

[0071] The present application will be further described in conjunction with the drawings and specific embodiments. As shown in the drawings, according to an embodiment of the present application, a cloud platform-based nursing robot motion prediction control method is provided, which comprises the following steps: Figure 1

[0072] S1, acquiring motion characteristic data of the nursing robot and the patient, wherein the motion characteristic data of the nursing robot and the patient comprises robot position data, robot speed data and patient posture data.

[0073] In one embodiment, acquiring the motion characteristic data of the nursing robot and the patient comprises the following steps:

[0074] S11, acquiring position data of a three-dimensional space of the robot, robot speed data and patient posture data, wherein the robot speed data comprises linear speed and angular speed;

[0075] For example, core sensors such as RGB-D cameras, IMU (Inertial Measurement Unit) sensors, force sensors, etc. are installed on the nursing robot to collect key data of the robot and the patient from different angles.

[0076] Position information: Obtain the three-dimensional space position information of the robot using a depth camera or an RGB-D camera.

[0077] Speed information: Obtain the linear speed and angular speed of the robot using an IMU sensor.

[0078] Posture information: Capture the posture of the robot and the patient in real time through the IMU and the RGB-D camera.

[0079] Collect environmental data at the same time: Obtain the obstacle position and scene changes of the surrounding environment through the depth camera.

[0080] S12, data filtering, normalization and time alignment are performed on the robot position data, robot speed data and patient posture data to ensure that the data is processed according to the same timeline.

[0081] ​S2, decompose the motion feature data of the nursing robot and the patient based on a singular value decomposition algorithm, and extract explicit and implicit features to obtain a comprehensive motion feature data set.

[0082] In one embodiment, decomposing the motion feature data of the nursing robot and the patient based on a singular value decomposition algorithm, and extracting explicit and implicit features to obtain a comprehensive motion feature data set includes the following steps:

[0083] S21, construct the robot position data, velocity data, and patient posture data into a multi-dimensional data matrix, wherein the rows represent different time points, and the columns represent different robot position data, robot velocity data, and patient posture data.

[0084] S22, perform singular value decomposition on the multi-dimensional data matrix using a singular value decomposition algorithm to decompose left singular vectors (U), right singular vectors (V), and singular values (Σ).

[0085] S23, extract explicit features (such as the main components of robot position, velocity, and patient posture) and implicit features (such as hidden patterns and trends in the data) from the decomposition results.

[0086] S24, re-integrate the explicit features and implicit features into a comprehensive motion feature data set.

[0087] In one embodiment, performing singular value decomposition on the multi-dimensional data matrix using a singular value decomposition algorithm to decompose left singular vectors, right singular vectors, and singular values includes the following steps:

[0088] S221, select a singular value decomposition algorithm (including standard SVD, Lanczos SVD, randomized SVD, and sparse SVD), and set the parameters for singular value decomposition, including the number of singular values to be decomposed;

[0089] wherein the standard SVD is suitable for small and medium-sized dense matrices.

[0090] Lanczos SVD: suitable for large dense matrices, and uses Lanczos algorithm for efficient calculation.

[0091] Randomized SVD: suitable for super large scale data sets, combined with random projection and iterative method, can efficiently process high-dimensional data.

[0092] Sparse SVD: suitable for sparse matrices, and optimized performance through sparse matrix calculation.

[0093] To determine the number of singular values, a rough number of singular values is initially set (e.g., according to the rank of the matrix or an empirical value). The data matrix is divided into a training set and a validation set (e.g., 80% training and 20% validation). Singular value decomposition is performed on the training set to obtain results of different numbers of singular values (e.g., k = 10, 20, 30,...). The reconstruction error (e.g., mean square error) is calculated for each decomposition. The performance of different numbers of singular values is evaluated using the validation set. The reconstruction error and the validation error are analyzed to detect whether overfitting or underfitting occurs. The number of singular values that best balances the reconstruction error and the validation error is selected. It is ensured that the selected number of singular values is stable under different data divisions.

[0094] S222, performing singular value decomposition on the multi-dimensional data matrix to obtain left singular vectors, singular values, and right singular vectors;

[0095] The left singular vectors (representing time characteristics) are used to reveal the change patterns of the nursing robot and the patient over time.

[0096] The singular values measure the importance of each feature and the corresponding time pattern, and the greater the singular value, the greater the influence of the corresponding feature on the overall motion.

[0097] The right singular vectors (weights of each feature) include the contribution sizes of the robot position data, the robot speed data, and the patient posture data, and are used to identify the influence of each feature in the motion pattern.

[0098] The singular values are diagonal elements of the decomposed matrix, representing the importance of each feature. The singular values provide a global perspective, measuring the contribution of each feature to the data structure, and are a quantitative indicator.

[0099] The right singular vectors represent the directions and weights of each feature vector, specifically including the contribution sizes of the robot position data, the speed data, and the patient posture data in the singular value decomposition. The right singular vectors provide a local perspective, explaining the relative contribution of specific features in forming the principal components.

[0100] In one embodiment, extracting explicit features (e.g., principal components of robot position, speed, and patient posture) and implicit features (e.g., implicit patterns and trends in the data) from the decomposition results includes the following steps:

[0101] S231, by analyzing the singular values and right singular vectors, the principal components of the robot position data, robot velocity data and patient posture data are identified, and the dominant features are obtained. The dominant features reflect the trajectory of the nursing robot movement, the change of the speed and the significant change of the patient posture, such as the displacement and acceleration of the robot when performing nursing tasks and the posture adjustment of the patient when receiving nursing. The dominant features of the robot position can reflect its main moving path, and the dominant features of the speed data can reveal the regularity of the robot acceleration or deceleration, and the dominant features of the patient posture reflect the significant posture adjustment of the patient.

[0102] S232, the dominant features are sorted according to the size of the singular values, and several dominant features are selected, that is, the dominant features with large singular values are preferentially selected.

[0103] S233, by analyzing the singular values and right singular vectors, the patterns and trends of the robot and the patient hidden in the data are extracted, and the latent features are obtained. The latent features are obtained by further analyzing the singular values and right singular vectors, and extracting the patterns and trends hidden in the data. These latent features include subtle coordination between the robot and the patient in interaction, or small movements of the patient when the posture is not obviously changed. For example, the latent features reveal the coordination between the robot and the patient position in subtle adjustment, or reflect the potential change trend in the patient posture, which are not easy to be observed by direct observation.

[0104] The dominant features are mainly used for key factors in explicit motion control, such as the moving direction of the robot, the speed change, and the obvious posture change of the patient. The latent features are used to capture the potential patterns in the motion, such as subtle posture changes and implicit interaction coordination, to provide deeper information support for motion prediction and control.

[0105] In one embodiment, by analyzing the singular values and right singular vectors, the patterns and trends hidden in the data are extracted, and the latent features include the following steps:

[0106] Based on the size of the singular values, the right singular vectors are reduced in dimension; the clustering algorithm is used to cluster the right singular vectors to identify different patterns and trends; the left singular vectors are analyzed by time series analysis to identify the latent feature patterns changing over time; for example, in each time window, the latent feature patterns are identified. Analyze the latent feature trends in the time series to identify long-term and short-term change rules; combine the clustering results with the time series analysis to generate the latent features.

[0107] S3, through the Internet of Things gateway, upload the comprehensive motion feature data set to the cloud platform, and train the motion prediction model based on the comprehensive motion feature data set in the cloud platform.

[0108] In one embodiment, uploading the comprehensive motion feature dataset to the cloud platform through the Internet of Things gateway, and training the motion prediction model based on the comprehensive motion feature dataset in the cloud platform includes the following steps:

[0109] S31, uploading the encrypted comprehensive motion feature dataset to the cloud platform through the Internet of Things gateway, decrypting the received comprehensive motion feature dataset in the cloud platform, and storing it in the database of the cloud platform.

[0110] S32, obtaining the label data corresponding to the comprehensive motion feature dataset, encoding the label data, and selecting a multi-layer perception for motion prediction.

[0111] S33, training and evaluating the multi-layer perception using the comprehensive motion feature dataset containing the label data to obtain the motion prediction model; specifically, training the multi-layer perception model using the training set data, including setting the number of hidden layers, the number of neurons, the activation function, the loss function, and the optimizer, etc. Adjust the hyperparameters of the model such as learning rate, batch size, training rounds, etc. to improve the performance of the model.

[0112] S34, packaging the trained motion prediction model and transmitting it to the local system of the nursing robot.

[0113] The encrypted model file is uploaded from the cloud platform to the local system of the nursing robot through the Internet of Things (IoT) gateway via a secure transmission protocol such as TLS or SSL. After the nursing robot receives the model file, it decrypts it and verifies the integrity and correctness of the file to ensure that the model has not been damaged or maliciously tampered with during transmission. To ensure the persistence of the model locally on the nursing robot, the model file will be securely stored in the local storage device

[0114] The loss function of the motion prediction model is:

[0115]

[0116] where x i represents the i-th motion data of the nursing robot predicted, y i represents the i-th motion data of the nursing robot actually occurred; m represents the number of training samples.

[0117] In one embodiment, the way to obtain the label data corresponding to the comprehensive motion feature dataset includes expert annotation method, sensor data fusion method, simulation or actual scene recording method and user feedback method.

[0118] Among them, the expert annotation method: according to the comprehensive motion feature dataset, the domain experts manually annotate the corresponding motion results as labels.

[0119] Sensor data fusion method: Collect data using various sensors on the nursing robot (such as accelerometers, gyroscopes, etc.), and obtain motion states as labels.

[0120] Simulated or actual scenario recording method: Record the motion of the nursing robot in a simulated or actual use scenario, and then extract motion features and corresponding motion results as training data.

[0121] User feedback method: During actual use, adjust and optimize the label data through user feedback on the motion of the nursing robot.

[0122] S4, output the next step motion of the nursing robot according to the motion prediction model, and the next step motion of the nursing robot includes trajectory data and pose data.

[0123] In one embodiment, outputting the next step motion of the nursing robot according to the motion prediction model, and the next step motion of the nursing robot includes trajectory data and pose data includes the following steps:

[0124] S41, obtain the latest comprehensive motion feature data set, and extract the comprehensive motion feature data at the current time as input data. That is, the latest motion feature information of the nursing robot and the patient, such as position, speed and pose, is converted into the current comprehensive motion feature data set.

[0125] S42, substitute the input data into the trained motion prediction model to calculate the next step motion prediction result of the nursing robot, including trajectory data and pose data.

[0126] S5, based on the output motion trajectory and pose data, generate corresponding control instructions to realize the motion control of the nursing robot.

[0127] In one embodiment, based on the output motion trajectory and pose data, generate corresponding control instructions to realize the motion control of the nursing robot includes the following steps:

[0128] S51, generate corresponding robot control instructions based on the predicted motion trajectory and pose data, which include how the robot moves, adjusts the pose, etc.

[0129] S52, combine environmental data for path planning to ensure that the nursing robot avoids obstacles during execution; for example, combine environmental data and predicted motion trajectory, and use path planning algorithms (such as A* algorithm, Dijkstra algorithm, RRT, etc.) to generate a safe path that avoids obstacles.

[0130] S53. Before executing control commands, collision detection and redundancy checks are performed on the generated robot control commands to ensure safety. Collision detection is achieved by simulating the robot's movement on the predicted path. Redundancy checks remove unnecessary duplicate commands, optimizing the efficiency and execution speed of the command set.

[0131] like Figure 2 As shown, according to another embodiment of the present invention, a cloud-based motion prediction and control system for nursing robots is also provided. The cloud-based motion prediction and control system for nursing robots includes a data acquisition module 1, a data decomposition module 2, a model training module 3, a motion prediction module 4, and a robot control module 5; the data acquisition module 1, the data decomposition module 2, the model training module 3, the motion prediction module 4, and the robot control module 5 are sequentially connected.

[0132] The system comprises five modules: a data acquisition module 1, and a robot control module 5. The data acquisition module 1 acquires motion characteristic data of the nursing robot and the patient, including robot position data, robot speed data, and patient posture data. The data decomposition module 2 decomposes the motion characteristic data of the nursing robot and the patient using a singular value decomposition algorithm, extracting explicit and implicit features to obtain a comprehensive motion characteristic dataset. The model training module 3 uploads the comprehensive motion characteristic dataset to a cloud platform via an IoT gateway and trains a motion prediction model based on the dataset within the cloud platform. The motion prediction module 4 outputs the next motion of the nursing robot based on the motion prediction model, including trajectory data and posture data. The robot control module 5 generates corresponding control commands based on the output motion trajectory and posture data to achieve motion control of the nursing robot.

[0133] To facilitate understanding of the above technical solutions of the present invention, the working principle or operation method of the present invention in actual process will be described in detail below.

[0134] 1. Acquire motion characteristic data of nursing robots and patients.

[0135] 1.1 Data Acquisition Equipment Installation and Initialization:

[0136] Install an RGB-D camera (to acquire 3D position information for the robot and patient), an inertial measurement unit (IMU) sensor (to acquire velocity and attitude information), and a force sensor. Initialize the sensors and confirm that all devices are connected to the data acquisition module and can acquire data in real time.

[0137] 1.2 Data Acquisition and Preprocessing:

[0138] Data collection from the robot and the patient:

[0139] Obtain the position data of the nursing robot (e.g. Pos1, Pos2,..., PosN, each position data records the three-dimensional position of the robot at different time points).

[0140] Collect the robot speed data (linear speed V1, V2,..., VN and angular speed A1, A2,..., AN).

[0141] Obtain the posture data of the patient (e.g. changes in the posture of the torso, limbs Pose1, Pose2,..., PoseN).

[0142] Data preprocessing:

[0143] Data filtering: use a low-pass filter to filter noise (F1), remove noise caused by sensor errors during the collection process.

[0144] Normalization: normalize the position, speed and posture data to the same numerical range, which is convenient for subsequent processing (N1).

[0145] Time alignment: align the data of different sensors according to the time stamp, ensuring that the data is aligned in the same time dimension (T1).

[0146] 1.3 Data storage:

[0147] Construct a data matrix (D1) from the filtered and normalized data, each row representing a different time point, and the columns representing the robot position, speed and patient posture characteristics. Input the motion characteristic data matrix (D1) of the nursing robot and the patient into the data decomposition module, and construct the data matrix:

[0148]

[0149] 2. Data decomposition and feature extraction

[0150] Select the singular value decomposition (SVD) algorithm (standard SVD, Lanczos SVD, etc.) and set the decomposition parameters, such as the number of singular values k for decomposition.

[0151] Perform SVD to obtain the left singular vector U, singular value Σ and right singular vector V: M = UΣV T

[0152] 2.3 Extract explicit and implicit features:

[0153] Dominant feature extraction: Extract features related to large singular values from the right singular vectors, such as major robot position, velocity, and attitude changes. Filter out important features and rank them by singular values. Latent feature extraction: Analyze the smaller singular value part to identify hidden motion patterns, such as subtle patient attitude changes and underlying trends in robot motion coordination. Cluster the right singular vectors using clustering algorithms to obtain different patterns and trends. Finally, generate a comprehensive motion feature dataset.

[0154] 3. Upload to cloud platform and train model

[0155] On the cloud platform, select an appropriate multi-layer perceptron (MLP) model and set the model structure (hidden layers, number of neurons, activation function, loss function, etc.). Train the model using the motion feature dataset and label data, adjusting hyperparameters (learning rate, batch size, etc.) to improve the model's prediction accuracy. Package and encrypt the trained model and securely transfer it to the nursing robot's local system using the Internet of Things gateway.

[0156] 4. Nursing robot motion prediction

[0157] 4.1 Obtain the latest motion feature data (D5)

[0158] Real-time data acquisition:

[0159] Collect patient's attitude data (Pose1) in real time through RGB-D camera, identify whether the patient is currently lying on his side, back or other posture. For example, the system identifies that the patient is currently in a supine position (Pose1: supine, angle 0 degrees).

[0160] Use IMU sensor to obtain the current position and velocity data (Pos1 and V1) of the nursing robot, for example, the robot is 50 cm away from the bed, and the speed is 0 (static state).

[0161] Real-time acquisition of surrounding obstacle information by environmental sensors (such as depth cameras) to ensure that the robot does not touch the bedside table or other objects during the turning process.

[0162] Construct a comprehensive motion feature dataset: D5 = {Pos1: (x = 50 cm, y = 0), V1: 0, Pose1: supine}

[0163] 4.2 Predict the next step motion

[0164] Input the above comprehensive data into the trained motion prediction model, which has been trained according to a large number of turning scenarios. Output the prediction result:

[0165] Motion trajectory: the robot needs to move 15 cm to the left side of the bed and adjust the angle to grab the patient's shoulder.

[0166] Pose data: The robot needs to extend its arm and adjust the joints at an angle to adapt to the grasping pose.

[0167] 5. Generate control instructions and execute

[0168] 5.1 Generate control instructions:

[0169] According to the predicted motion trajectory and pose data, specific control instructions are generated, such as:

[0170] The robot moves 15 cm to the left and rotates 20 degrees.

[0171] The robot moves to the specified position and adjusts the joints to 45 degrees to grasp the patient's shoulder.

[0172] 5.2 Path planning and safety detection

[0173] Path planning: Combine environmental sensor data and use path planning algorithms to plan a safe path to avoid collisions between the robot and obstacles during movement. For example, path planning may show that the care robot needs to avoid obstacles at the bedside during the 15 cm movement.

[0174] Collision detection: Before executing the control instructions, the system will simulate whether there is a collision risk on the path and check whether the instructions are redundant or unnecessary, to ensure that the robot performs actions with the optimal path.

[0175] After the safety detection confirms that there is no conflict in the instructions, the control instructions can be updated to: {no collision risk, executable instructions}.

[0176] In summary, the present application collects multi-dimensional motion feature data of the robot and the patient, and performs dimensionality reduction processing on the multi-dimensional motion feature data through singular value decomposition, to remove noise and redundant information, improve data quality and processing efficiency. Extracting explicit and implicit features, identifying the main patterns and potential relationships in the data, provides a more representative and generalizable feature set for the motion prediction model. Data decomposition and feature extraction can compress data volume without losing key information, reducing the burden of subsequent calculation and storage. Through the Internet of Things gateway, efficient data transmission is realized, ensuring that data can be timely uploaded to the cloud platform for processing. Using the powerful computing power and resources of the cloud platform, complex motion prediction model training is carried out, improving the accuracy and stability of the model. The cloud platform can realize continuous data update and model optimization, so that the motion prediction model can continuously learn and adapt to new data and changes. Based on the trained motion prediction model, the next step of the care robot can be accurately predicted, i.e. the robot's motion is planned in advance, reducing the occurrence of unexpected situations.

[0177] The above merely provides the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A cloud platform-based nursing robot motion prediction control method, characterized in that, The cloud platform-based nursing robot motion prediction control method comprises the following steps: S1, obtain the motion characteristic data of the nursing robot and the patient, and the motion characteristic data of the nursing robot and the patient comprises robot position data, robot speed data and patient posture data; S2, decompose the motion characteristic data of the nursing robot and the patient based on a singular value decomposition algorithm, and extract explicit and implicit features to obtain a comprehensive motion characteristic data set; S3, upload the comprehensive motion characteristic data set to the cloud platform through the Internet of Things gateway, and train a motion prediction model based on the comprehensive motion characteristic data set in the cloud platform; S4, output the next step motion of the nursing robot according to the motion prediction model, and the next step motion of the nursing robot comprises trajectory data and posture data; S5, generate corresponding control instructions based on the output motion trajectory and posture data to realize the motion control of the nursing robot; Wherein, the singular value decomposition algorithm is used to decompose the motion characteristic data of the nursing robot and the patient, and extract explicit and implicit features to obtain a comprehensive motion characteristic data set, which comprises the following steps: S21, construct the robot position data, speed data and patient posture data into a multi-dimensional data matrix, wherein the rows represent different time points, and the columns represent different robot position data, robot speed data and patient posture data; S22, perform singular value decomposition on the multi-dimensional data matrix using a singular value decomposition algorithm to decompose left singular vectors, right singular vectors and singular values; S23, extract explicit and implicit features from the decomposition results; S24, re-integrate the explicit and implicit features into a comprehensive motion characteristic data set. 2.The cloud platform-based nursing robot motion prediction control method of claim 1, wherein, The motion characteristic data of the nursing robot and the patient is obtained by the following steps: S11, obtain the robot position data, robot speed data and patient posture data, and the robot speed data comprises linear speed and angular speed; S12, perform data filtering, normalization and time alignment on the robot position data, robot speed data and patient posture data. 3.The cloud platform-based nursing robot motion prediction control method of claim 1, wherein, The singular value decomposition algorithm is used to perform singular value decomposition on the multi-dimensional data matrix to decompose left singular vectors, right singular vectors and singular values, which comprises the following steps: S221, select a singular value decomposition algorithm and set the parameters of singular value decomposition, including the number of singular values to be decomposed; S222, perform singular value decomposition on the multi-dimensional data matrix to obtain left singular vectors, singular values and right singular vectors; The left singular vectors are used to reveal the change pattern of the motion of the nursing robot and the patient over time; The singular values are used to measure the importance of each feature and the corresponding time pattern, and the greater the singular value, the greater the influence of the corresponding feature on the overall motion; The right singular vectors comprise the contribution size of the robot position data, robot speed data and patient posture data, and are used to identify the influence degree of each feature in the motion pattern. 4.The cloud platform-based nursing robot motion prediction control method of claim 3, wherein, The explicit and implicit features are extracted from the decomposition results by the following steps: S231, identify the principal components of the robot position data, robot speed data and patient posture data by analyzing the singular values and right singular vectors to obtain explicit features; S232, rank the explicit features according to the size of the singular values, and select a number of explicit features; S233, extract the patterns and trends of the robot and the patient hidden in the data by analyzing the singular values and the right singular vectors, to obtain implicit features. 5.The cloud platform-based nursing robot motion prediction control method of claim 4, wherein, The step of extracting the patterns and trends hidden in the data by analyzing the singular values and the right singular vectors to obtain implicit features includes the following steps: S2331, dimensionality reduction of the right singular vector based on the size of the singular value; S2332, clustering the right singular vector using a clustering algorithm to identify different patterns and trends; S2333, time series analysis of the left singular vector to identify the implicit feature patterns that change over time; S2334, combine the clustering results in step S2332 with the time series analysis in step S2333 to generate implicit features. 6.The cloud platform-based nursing robot motion prediction control method of claim 1, wherein, The step of uploading the comprehensive motion feature dataset to the cloud platform through the Internet of Things gateway, and training a motion prediction model based on the comprehensive motion feature dataset in the cloud platform includes the following steps: S31, upload the encrypted comprehensive motion feature dataset to the cloud platform through the Internet of Things gateway, decrypt the received comprehensive motion feature dataset in the cloud platform, and store it in the database of the cloud platform; S32, obtain the label data corresponding to the comprehensive motion feature dataset, encode the label data, and select a multi-layer perceptron for motion prediction; S33, train and evaluate the multi-layer perceptron using the comprehensive motion feature dataset containing the label data to obtain a motion prediction model; S34, package the trained motion prediction model and transmit it to the local system of the nursing robot; The loss function of the motion prediction model is: In the formula, x i represents the i-th motion data of the predicted nursing robot, y i represents the i-th motion data of the actually occurred nursing robot; m represents the number of training samples. 7.The cloud platform-based nursing robot motion prediction control method of claim 6, wherein, The way to obtain the label data corresponding to the comprehensive motion feature dataset includes expert annotation, sensor data fusion, scene recording, and user feedback. 8.The cloud platform-based nursing robot motion prediction control method of claim 1, wherein, The step of outputting the next motion of the nursing robot according to the motion prediction model, and the next motion of the nursing robot including trajectory data and posture data includes the following steps: S41, obtain the latest comprehensive motion feature dataset and extract the comprehensive motion feature data at the current time as input data; S42, substitute the input data into the trained motion prediction model to calculate the motion prediction result of the next step of the nursing robot, including trajectory data and posture data. 9.The cloud platform-based nursing robot motion prediction control method of claim 8, wherein, The step of generating corresponding control instructions based on the output motion trajectory and posture data to realize the motion control of the nursing robot includes the following steps: S51, generate corresponding robot control instructions based on the predicted motion trajectory and posture data; S52, combine environmental data for path planning to ensure that the nursing robot avoids obstacles during execution; S53, perform collision detection and redundancy check on the generated robot control instructions to ensure safety.

10. A cloud platform-based nursing robot motion predictive control system for implementing the cloud platform-based nursing robot motion predictive control method of any one of claims 1-9, characterized in that, The cloud platform-based nursing robot motion prediction control system includes a data acquisition module, a data decomposition module, a model training module, a motion prediction module, and a robot control module; The data acquisition module, the data decomposition module, the model training module, the motion prediction module and the robot control module are sequentially connected. The data acquisition module is configured to acquire motion characteristic data of the nursing robot and the patient, and the motion characteristic data of the nursing robot and the patient includes robot position data, robot speed data and patient posture data. The data decomposition module is configured to decompose the motion characteristic data of the nursing robot and the patient based on a singular value decomposition algorithm, and extract explicit and implicit features to obtain a comprehensive motion characteristic data set. The model training module is configured to upload the comprehensive motion characteristic data set to a cloud platform through an Internet of Things gateway, and train a motion prediction model based on the comprehensive motion characteristic data set in the cloud platform. The motion prediction module is configured to output the next step motion of the nursing robot according to the motion prediction model, and the next step motion of the nursing robot includes trajectory data and posture data. The robot control module is configured to generate corresponding control instructions based on the output motion trajectory and posture data to realize motion control of the nursing robot.

Images