A control method and system of a care robot

Abstract

The present application relates to the technical field of nursing robot control, and especially relates to a control method and system of a nursing robot. The method comprises the following steps: using a 3D camera to collect the posture trajectory of a patient when getting up, and using an electromyographic signal sensor to obtain a muscle tension change signal associated therewith, and obtaining posture-associated muscle tension data through synchronous correlation processing. Then, the data is analyzed, the muscle group force line imbalance strength is evaluated, and the center of gravity offset trend is obtained by combining the center of gravity offset regression analysis. Finally, the robot auxiliary output parameters are generated based on the center of gravity offset regression trend, input into the nursing robot control architecture, a control scheme suitable for the nursing robot is designed, and the control instructions are sent to the robot terminal to perform corresponding nursing operations. The present application optimizes the control technology of the nursing robot, and makes the control technology of the nursing robot more perfect.

Description

Technical Field

[0001] This invention relates to the field of nursing robot control technology, and in particular to a control method and system for a nursing robot. Background Technology

[0002] Existing nursing robots primarily rely on simple programmed control and basic sensor feedback to function; however, this approach has limitations. For example, during patient standing or positional changes, traditional robots cannot accurately perceive the patient's real-time physical condition and needs, resulting in imprecise robot control and an inability to adjust promptly based on the patient's specific state, potentially even causing discomfort or injury. Therefore, intelligent nursing robots need higher-level control capabilities, not only sensing changes in patient posture but also analyzing physiological signals such as muscle tension and center of gravity shifts in real time to more accurately assist patient movement. However, traditional nursing robot control methods often fail to accurately perceive changes in the patient's center of gravity and muscle tension fluctuations during standing, leading to significant control errors during assisted movement. Summary of the Invention

[0003] Therefore, it is necessary to provide a control method and system for a nursing robot to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a control method for a nursing robot is provided, the method comprising the following steps:

[0005] Step S1: Acquire the patient's posture trajectory when standing up during the nursing process using a 3D camera; acquire muscle tension change signals in the body contact area when the patient stands up using an electromyography (EMG) sensor based on the posture trajectory; perform synchronous correlation processing on the muscle tension change signals based on the posture trajectory to obtain posture-related muscle tension change data.

[0006] Step S2: Analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data. Then, perform center of gravity offset regression deconstruction to obtain the center of gravity offset regression trend. Generate robot-aided output parameters based on the center of gravity offset regression trend.

[0007] Step S3: Use the robot-assisted output parameters as input to design the control architecture of the nursing robot, so as to build a robot-assisted nursing architecture, and send it to the control terminal of the nursing robot to execute the control of the nursing robot.

[0008] Preferably, the present invention also provides a control system for a nursing robot, used to execute the control method for the nursing robot described above, the control system comprising:

[0009] The data acquisition module is used to acquire the patient's posture trajectory when getting up during the nursing process using a 3D camera; and to acquire the muscle tension change signals in the body contact area when the patient gets up during the nursing process using an electromyography (EMG) sensor based on the posture trajectory; and to perform synchronous correlation processing on the muscle tension change signals based on the posture trajectory to obtain posture-related muscle tension change data.

[0010] The robot output parameter analysis module is used to analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data. Subsequently, it performs center of gravity offset regression deconstruction to obtain the center of gravity offset regression trend; and generates robot auxiliary output parameters based on the center of gravity offset regression trend.

[0011] The control architecture design module is used to design the control architecture of the nursing robot by taking the robot-assisted output parameters as input, so as to build the robot-assisted nursing architecture and send it to the control terminal of the nursing robot to execute the control of the nursing robot.

[0012] The beneficial effects of this invention lie in its ability to collect real-time data on a patient's posture trajectory and muscle tension changes during standing up, achieved through the combination of a 3D camera and an electromyography (EMG) sensor. This process ensures comprehensive perception of the patient's physical condition, particularly during dynamic changes, accurately capturing subtle changes in each movement. By synchronously processing this data, high-precision posture-related muscle tension change data can be generated, providing a reliable foundation for subsequent data analysis and control decisions. This high-precision data acquisition and processing method enables the robot to respond precisely to the patient's actual needs, improving the intelligence and personalization of nursing services. In-depth analysis of the posture-related muscle tension change data assesses the intensity of muscle group force line imbalance during posture changes. This analysis reveals the biomechanical state of various parts of the patient's body during standing up, promptly identifying postural imbalances or maladaptive behaviors. Further regression analysis of center of gravity shifts reveals the trend of the patient's center of gravity changes during standing up, accurately predicting the stability and risks during movement. This process provides crucial physiological data support, allowing the robot to dynamically adjust based on the patient's posture and center of gravity changes, ensuring safety and comfort during movements such as standing up. Based on the robot-assisted output parameters generated in the first two steps, an adaptive nursing robot control architecture was designed. This architecture can dynamically control the robot according to the patient's real-time physiological state (such as posture changes, muscle tension, center of gravity shift, etc.). By receiving these output parameters, the robot adjusts its movements to adapt to the individual needs of the patient, precisely assisting the patient in completing actions such as getting up and adjusting body position. This control architecture enables the robot to provide precise assistance in complex nursing scenarios, avoiding patient discomfort or injury caused by mechanized movements. Through this adaptive control, the nursing robot not only improves the safety of nursing services but also enhances nursing efficiency and quality, enabling its application in a wider range of clinical nursing scenarios. Therefore, this invention is an optimization of a traditional nursing robot control method, solving the problem that traditional nursing robot control methods cannot accurately perceive changes in the center of gravity and muscle tension fluctuations when the patient gets up during the nursing process, resulting in large control errors during assisted assistance. This invention improves the accuracy of perceiving changes in the center of gravity and muscle tension fluctuations when the patient gets up during the nursing process, reducing the control errors of the nursing robot during assisted assistance. Attached Figure Description

[0013] Figure 1 A flowchart illustrating the steps of a control method for a nursing robot;

[0014] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0015] Figure 3 for Figure 2 A detailed flowchart illustrating the implementation steps of step S22. Detailed Implementation

[0016] Please see Figures 1 to 3 A method for controlling a nursing robot, the method comprising the following steps:

[0017] Step S1: Acquire the patient's posture trajectory when standing up during the nursing process using a 3D camera; acquire muscle tension change signals in the body contact area when the patient stands up using an electromyography (EMG) sensor based on the posture trajectory; perform synchronous correlation processing on the muscle tension change signals based on the posture trajectory to obtain posture-related muscle tension change data.

[0018] Step S2: Analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data. Then, perform center of gravity offset regression deconstruction to obtain the center of gravity offset regression trend. Generate robot-aided output parameters based on the center of gravity offset regression trend.

[0019] Step S3: Use the robot-assisted output parameters as input to design the control architecture of the nursing robot, so as to build a robot-assisted nursing architecture, and send it to the control terminal of the nursing robot to execute the control of the nursing robot.

[0020] In this embodiment of the invention, reference is made to Figure 1 The above is a flowchart illustrating the steps of a control method for a nursing robot according to the present invention. In this example, the control method for the nursing robot includes the following steps:

[0021] Step S1: Acquire the patient's posture trajectory when standing up during the nursing process using a 3D camera; acquire muscle tension change signals in the body contact area when the patient stands up using an electromyography (EMG) sensor based on the posture trajectory; perform synchronous correlation processing on the muscle tension change signals based on the posture trajectory to obtain posture-related muscle tension change data.

[0022] In this embodiment of the invention, a three-dimensional camera is installed on the head of the nursing robot. Based on the principle of structured light imaging, this camera continuously captures spatial point cloud data at a frame rate of 30Hz. Each frame contains 200,000 three-dimensional points, and the measurement range is set within a spatial area of ​​1.2m × 0.8m × 1.5m. This range covers the entire movement trajectory of the patient from a sitting to an upright position at the edge of the nursing bed. Simultaneously, an integrated electromyography (EMG) sensor array is adhered to the inner contact surface of the robotic arm's palm. Each sensor array consists of four electrodes with a 20mm spacing. Each electrode point corresponds to a muscle group at the patient's contact point, including the forearm flexor muscles, levator scapulae muscles, and triceps brachii muscles. When the patient stands up, their palm, forearm, and back contact the robotic arm's palm, and the electrodes directly sense the surface EMG potential. The acquisition process uses a 16-bit analog-to-digital converter to record EMG signals at a sampling frequency of 1000Hz. Before entering the buffer, the signal is processed by a bandpass filter with a lower cutoff frequency limit of 0.5Hz and an upper cutoff frequency limit of 200Hz. A notch filter is also configured to eliminate 50Hz power frequency noise. The processed sequence is then timestamped with millisecond precision, and the timestamps are derived from the camera's internal clock. Subsequently, the electromyographic (EMG) data and posture data are synchronized and matched according to the frame number of the spatial trajectory captured by the camera. For example, at 3.25s, the camera records the trunk center of mass position as (0.38m, 0.22m, 0.90m), and simultaneously, the forearm EMG electrode output voltage amplitude is 0.54V, and the scapular electrode output voltage amplitude is 0.72V. The data at this time point is correlated into posture-tension one-to-one corresponding data units, ultimately forming complete posture-related muscle tension change data.

[0023] Step S2: Analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data. Then, perform center of gravity offset regression deconstruction to obtain the center of gravity offset regression trend. Generate robot-aided output parameters based on the center of gravity offset regression trend.

[0024] In this embodiment of the invention, after forming posture-tension correlation data, the spatial coordinate reconstruction of the patient's standing process is first required. The reconstruction employs a human keypoint extraction method, extracting the coordinates of the hip, knee, ankle, shoulder, elbow, and lumbar spine center points from the 3D point cloud. For example, in frame 25, the coordinates of the left hip joint are (0.22m, 0.16m, 0.85m), the right hip joint is (0.21m, 0.15m, 0.85m), and the center of mass is (0.215m, 0.155m, 0.87m). Simultaneously, the electromyographic tension data is processed using first-order difference to obtain the muscle contraction rate; a rate higher than 0.05V / s is considered a significant tension change in the muscle. In this way, multiple sets of tension vectors are obtained during specific standing phases. Next, using the vertical direction of the torso as the reference axis, the tension change of each muscle group is decomposed into three components: the anterior-posterior direction (x-axis), the lateral direction (y-axis), and the vertical direction (z-axis). The resulting component data form a spatial force line distribution vector set. Subsequently, an orthogonal second-order decomposition is performed on this vector set to obtain the variance in each direction, thereby calculating the degree of imbalance in the muscle group force lines in different directions. For example, when the patient rises to 5.50 seconds, the tension in the left upper arm in the y-direction is 0.85N, and the tension in the right upper arm in the y-direction is 1.25N. The difference is 0.40N. By comparing this with the mean squared error of all data, the force line imbalance intensity is quantified as 35%. Next, this force line imbalance intensity is mapped onto the dynamic trajectory of the center of mass. The center of mass trajectory is calculated using a frame sequence; from crawling to fully upright, the center of mass rises by a total of 0.30m. The displacement curve of the entire ascent process is fitted using a third-order polynomial to form the regression trend of the body's center of mass trajectory. Difference is performed on the trend to obtain the changes in the center of mass offset in the lateral and longitudinal directions. For example, in the experiment, the lateral offset was 0.025m and the longitudinal offset was 0.30m. The upward trend was generally stable, but the lateral offset coupled with the force line imbalance intensity of 35%. This value was directly output as the key input information in the subsequent robot auxiliary output parameters.

[0025] In another embodiment, using the posture-related muscle tension change data generated in step S1, the auxiliary output parameters that the robot needs to execute are generated by analyzing the relationship between muscle force imbalance and the body's center of gravity shift. First, the muscle group force line imbalance intensity analysis is performed on the posture-related muscle tension change data. Since the electromyographic signals originate from the patient's back area contacted by the robot's hand, the analysis focuses on the synergistic force exertion pattern of the erector spinae and trapezius muscles in this area. The tension intensity values ​​of the 16 electromyographic channels are constructed into a 4x4 two-dimensional matrix, representing the muscle tension distribution map on the sensor contact surface. At each time point, the coordinates of the centroid of muscle activity intensity at that moment on the two-dimensional plane are obtained by weighted averaging of this matrix. The degree of deviation of this centroid coordinate from the sensor's geometric center is defined as the force line imbalance vector of the local muscle group. The magnitude of this vector is the force line imbalance intensity. Subsequently, a center of gravity shift regression deconstruction is performed. Using the 21 joint point 3D coordinates and standard human segment mass distribution parameters obtained in step S1, the overall 3D center of gravity coordinates of the patient's body in each frame are calculated by weighted summation, thus constructing a continuous dynamic center of gravity trajectory. Next, the intensity of force line imbalance at each moment is correlated with the horizontal movement velocity of the dynamic center of gravity at the same moment. Specifically, the "intensity of force line imbalance" in the continuous time series is used as the dependent variable, and the "horizontal velocity of the center of gravity" is used as the independent variable, to perform a univariate linear regression analysis. The purpose of this analysis is to establish a quantitative relationship, namely, to what extent the rapid movement of the center of gravity leads to the imbalance of the back supporting muscles. A regression coefficient is obtained after the regression analysis; this coefficient represents the center of gravity shift regression trend. For example, a calculated regression coefficient of 0.08 means that for every 0.1 m / s increase in the patient's horizontal velocity of the center of gravity, the center of mass of the back muscle activity deviates from the contact center by 8 mm. Finally, robot-assisted output parameters are generated based on this regression trend. A trigger threshold is set. When the real-time calculated force line imbalance intensity exceeds 0.015m (1.5cm) and the horizontal velocity of the center of gravity is greater than 0.25m, the system determines that the patient has an unstable tendency. The robot-assisted output parameters generated at this time are: an auxiliary force vector, whose direction is opposite to the direction of the force line imbalance vector, which aims to push the center of mass of the muscle activity back to the contact center; the magnitude of the auxiliary force is calculated by a proportional-integral controller, whose set value is 500 times the force line imbalance intensity. For example, when the imbalance intensity is 0.016m, the magnitude of the auxiliary force is 8N; the point of application of the auxiliary force is the robot's current contact point.

[0026] Step S3: Use the robot-assisted output parameters as input to design the control architecture of the nursing robot, so as to build a robot-assisted nursing architecture, and send it to the control terminal of the nursing robot to execute the control of the nursing robot.

[0027] In this embodiment of the invention, the auxiliary output parameters obtained in step S2 consist of two parts: a center of gravity shift regression trend sequence and the associated muscle group force line imbalance intensity value. These parameters are input to the central control unit of the nursing robot. The central control unit first performs parameter feature weight allocation. According to preset rules, the weight of the center of gravity shift trend is set to 0.6, and the weight of the muscle group force line imbalance intensity is set to 0.4. After feature learning, an auxiliary parameter instruction set is generated. The auxiliary parameter instruction set is divided into three elements: functional factors, temporal factors, and constraint factors. Functional factors include the robot arm's balance support, assisted lifting, and lateral stabilization; temporal factors require parameter instructions to be refreshed every 0.02s; constraint factors limit the single output force to no more than 30N, and the robot arm's angular velocity to no more than 0.05rad / s. Taking a real-world experiment as an example, when the patient's center of gravity shifts 0.025m to the left and their longitudinal displacement is 0.30m during the process of getting up, according to the auxiliary parameter instructions, the robot's right-side robotic arm actuator generates a supporting force of 9N, and the left-side robotic arm actuator generates a stabilizing force of 6N. The robot's hand contacts the patient's back and forearm, maintaining the contact force within the specified range. The drive control frequency is maintained at 50Hz to ensure process continuity. The control architecture transmits each set of multi-dimensional instructions to the end-effector joint actuators via a data bus, thereby achieving auxiliary support and dynamic coordination. In this way, the nursing robot's movements are precisely synchronized with the patient's motion trajectory throughout the entire process of getting up.

[0028] Step S1 includes the following steps:

[0029] Step S11: Capture the patient's posture trajectory when getting up during the nursing process using a 3D camera;

[0030] Step S12: Using an electromyography (EMG) sensor and based on the posture trajectory, collect muscle tension change signals in the body contact areas when the patient gets up during the nursing process; wherein the body contact areas include: hand / forearm area, back and scapula area and elbow / upper arm area;

[0031] Step S13: Filter the muscle tension change signal for noise and then embed the timestamp to obtain the time series of the change signal;

[0032] Step S14: Based on the posture trajectory, perform synchronous correlation processing on the time series of the change signal to obtain posture-related muscle tension change data.

[0033] In this embodiment of the invention, a 3D camera is fixedly mounted on the head of the nursing robot. The camera uses structured light to construct spatial point clouds at a frame rate of 30Hz, outputting 200,000 spatial points per frame. The acquisition range is fixed as a cubic space with a length of 1.2m, a width of 0.8m, and a height of 1.5m, used to cover the entire motion area from sitting to standing. During the experiment, the patient was placed beside the nursing bed, and the patient changed from a lying position to a sitting position and then gradually stood up. The entire process took about 6.5 seconds, and the camera was able to continuously capture 195 frames of motion data. The human skeleton key point recognition algorithm was called from the point cloud data to extract the coordinates of 29 joint points, including the hip joint, knee joint, ankle joint, shoulder joint, elbow joint, and the center point of the lumbar spine. The coordinates of each joint point are expressed in meters (m) in a three-dimensional Euclidean coordinate system. For example, at time 2.50s, the center of mass coordinates are (0.38m, 0.20m, 0.92m), the left knee joint coordinates are (0.20m, 0.32m, 0.53m), the right knee joint coordinates are (0.21m, 0.33m, 0.52m), and the torso angle is 15°. A sequence of posture trajectory data is formed through consecutive frames. The trajectory includes not only joint position coordinates but also the rate of change of each joint's angle relative to the vertical axis. This trajectory data provides a spatial reference for the temporal matching of subsequent muscle tension data.

[0034] A four-channel surface electromyography (EMG) sensor array was embedded in the palm region of the robotic arm of a nursing robot. Each channel consisted of two 10mm diameter solid-state electrodes with a 20mm electrode spacing, a sampling frequency of 1000Hz, and a resolution of 16-bit. The sensor array was fixed to the corresponding robotic contact surfaces of the patient's palm, forearm, elbow, and scapula. During the experiment, when the patient stood up and grasped the robotic arm, the hand electrodes measured the potential signals of the flexor muscle groups. Taking a real experiment as an example, during the patient's standing up at 2.80s, the peak voltage measured by the hand electrodes was 0.55V, the EMG amplitude at the elbow was 0.48V, and the amplitude of the scapular muscle group was 0.67V. The collected data were voltage-time series, in volt-seconds (V·s). To maintain matching with the posture trajectory, each set of acquisitions included a precise sampling timestamp with an error of less than 1ms. This EMG signal reflects the amplitude and intensity of tension changes in the muscle groups at each contact point during movement, providing a basic input for force line direction analysis and center of gravity shift calculation.

[0035] After data acquisition, the electromyographic (EMG) signals were preprocessed. A bandpass filter was used with parameters set to a lower limit of 0.5 Hz and an upper limit of 200 Hz to remove baseline drift and high-frequency interference. An additional notch filter (bandwidth ±2 Hz) was added to suppress 50 Hz interference from the power supply system. The filtered signal was then rectified and its envelope extracted to reflect the time-varying trend of EMG energy. In the experiment, from the time the patient sat up until 4.10 s, the amplitude of the filtered envelope signal from the scapular electrode remained at 0.75 V, the noise level was less than 0.05 V, and the signal-to-noise ratio reached 15:1. The processed signal data stream was embedded with a timestamp from a unified clock every millisecond. The timestamp was generated by a clock pulse shared by the camera and sensor, ensuring all data sequences were on the same time base. The final output was a paired sequence of "potential intensity-time". For example, at 4.00s, the hand signal amplitude is 0.48V, corresponding to a timestamp of 4000ms, and the back signal amplitude is 0.80V, also corresponding to a timestamp of 4000ms. This sequence lays the foundation for synchronization processing.

[0036] Under the timestamp synchronization mechanism, the posture trajectory data sampled from the 3D camera is aligned millisecond by millisecond with the processed electromyographic time series. For each timestamp, the corresponding centroid coordinates and local joint data are found and bound to the corresponding electrode potential amplitude. For example, at 3.50s, the centroid position is (0.36m, 0.24m, 0.95m), while the forearm electrode voltage is 0.60V and the scapular electrode voltage is 0.73V. The posture-tension correlation data at this moment is stored in the joint sequence. Through continuous synchronization, a complete posture-related muscle tension change data sequence is obtained. This dataset contains the spatial coordinates, joint angles, centroid offset, and electromyographic potential amplitude at each time point. In further processing, the data is interpolated to ensure that data gaps caused by the difference between the camera frame rate and the electromyographic sensor sampling rate are eliminated. Finally, a posture-tension joint dataset covering the entire 6.5s movement process is obtained, with an accuracy of 1ms and an alignment error of less than 0.002s. This implementation process lays the foundation for a one-to-one correspondence between time and space data for subsequent force line vector decomposition and centroid regression.

[0037] Step S2 includes the following steps:

[0038] Step S21: Construct the spatial coordinates of the posture trajectory during the standing process to obtain the spatial coordinates of the posture trajectory;

[0039] Step S22: Analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data to obtain the force line imbalance intensity.

[0040] Step S23: Based on the attitude trajectory spatial coordinates, perform a center of gravity offset regression deconstruction on the force line imbalance intensity to obtain the center of gravity offset regression trend;

[0041] Step S24: Based on the center of gravity offset regression trend, combined with the spatial coordinates of the posture trajectory and the posture-related muscle tension change data, generate robot-assisted output parameters.

[0042] As an example of the present invention, reference is made to... Figure 2 As shown, in this example, step S2 includes:

[0043] Step S21: Construct the spatial coordinates of the posture trajectory during the standing process to obtain the spatial coordinates of the posture trajectory;

[0044] In this embodiment of the invention, the patient's posture trajectory established in step S1 is used to construct spatial coordinates. A 3D camera outputs point cloud data in a continuous time series, acquiring 30 frames per second. Each frame of point cloud data is processed by a human joint detection algorithm to determine the 3D coordinates of the hip, knee, ankle, shoulder, elbow, and lumbar spine center points. For example, at 2.00s, the coordinates of the left hip joint are (0.22m, 0.18m, 0.82m), the right hip joint is (0.20m, 0.17m, 0.81m), and the centroid coordinates are (0.21m, 0.175m, 0.835m). Spatial coordinate construction is not merely a single coordinate record; it also forms a continuous trajectory curve through the joint point sequence. During the standing process, the speed and angle changes of each joint are also calculated through the difference between adjacent frames. Taking the right knee joint as an example, at 2.00s, the coordinates are (0.18m, 0.30m, 0.50m), and at 2.03s, the coordinates are (0.18m, 0.305m, 0.52m). The difference between these two points outputs a local velocity vector of (0.0m / s, 0.17m / s, 0.67m / s). By performing this operation on all joints, the dynamic coordinate matrix of the entire upper and lower body joints is obtained. This dynamic matrix contains not only absolute position but also rate of change information, ultimately forming a complete spatial coordinate sequence of the standing trajectory, which provides accurate spatial support for subsequent mechanical analysis.

[0045] Step S22: Analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data to obtain the force line imbalance intensity.

[0046] In this embodiment of the invention, after establishing the spatial coordinates, force line imbalance intensity analysis is performed based on the synchronous sequence of posture and muscle tension data. The collected electromyographic signals are filtered and converted into muscle group tension amplitudes. The electrical signal, measured in volts, has a linear relationship with muscle tension. Through experimental calibration, the electrical signal amplitude is mapped to an equivalent force value. For example, the voltage amplitude of the forearm electrode at 3.00s is 0.55V, and the stress value is calibrated to 10.2N. The force vector of each contact area muscle group is decomposed according to the body's three-dimensional coordinate system to obtain components along the x, y, and z directions. For example, the force corresponding to the back muscle group at 3.20s is 15N, with the decomposition results being 4.0N in the x direction, 3.5N in the y direction, and 7.5N in the z direction. Then, the data of the left and right symmetrical muscle groups are calculated by difference. When the difference between the two symmetrical parts is higher than a set threshold of 5N, it is defined as force line imbalance. For example, the difference between the left and right arms in the y direction is 6.2N, which is higher than the threshold and is judged to be imbalanced. Throughout the entire standing motion, frame-by-frame statistical data of imbalance intensity is used to form a time curve. For example, the maximum imbalance intensity is 35% in the range of 3.00s to 4.50s. This process achieves a measurement of the mechanical inhomogeneity of muscle groups during posture changes, and the resulting time series serves as input for subsequent center of gravity trend regression.

[0047] Step S23: Based on the attitude trajectory spatial coordinates, perform a center of gravity offset regression deconstruction on the force line imbalance intensity to obtain the center of gravity offset regression trend;

[0048] In this embodiment of the invention, the patient's center of mass trajectory is calculated using spatial coordinates. The center of mass position is obtained by weighted averaging of key points throughout the body. During the ascent, the center of mass rises from coordinates (0.21m, 0.175m, 0.835m) to (0.22m, 0.23m, 1.12m), with a longitudinal shift of 0.285m and a lateral shift of 0.01m. A third-order polynomial is fitted to this center of mass displacement curve, and the fitted curve represents the continuous trend of the center of mass change. To accurately reflect the influence of muscle force imbalance, the force line imbalance intensity sequence is coupled with the center of mass trajectory analysis. At each moment, the imbalance intensity is mapped to the corresponding center of mass coordinate point, forming a binary relationship of "local imbalance - center of mass position". For example, at 3.50s, the imbalance intensity is 28%, and the center of mass position is (0.22m, 0.20m, 0.98m). This mapping relationship is then differentially analyzed to calculate the small shift trend of the center of mass with the imbalance intensity, thereby deconstructing the center of mass shift regression trend. During the experiment, the longitudinal acceleration was measured to be 0.20 m / s² during the period from 4.00 s to 4.30 s. The corresponding lateral imbalance resulted in a lateral shift of the center of gravity of 0.025 m. This value is the result of the center of gravity shift regression analysis, providing a quantitative basis for the generation of auxiliary parameters in the next step.

[0049] In another embodiment, a quantitative relationship is established between the intensity of force line imbalance and the dynamic changes in the patient's center of gravity, thereby deconstructing the trend of center of gravity shift. First, based on the attitude trajectory spatial coordinates obtained in step S21, the overall three-dimensional center of gravity coordinates of the patient's body in each frame are calculated. This calculation is performed using a set of anthropometric parameters comprising 15 body segments (e.g., head, trunk, left and right thighs, lower legs, etc.). This parameter set defines the percentage of mass of each segment relative to the total body weight (e.g., 49.7% for the trunk) and the relative position of its center of mass within that segment. The coordinates of the center of mass of the entire body are obtained by taking a mass-weighted average of the coordinates of the centers of mass of all segments. Next, the velocity of the center of mass offset in the horizontal plane (XY plane) is calculated. The calculation method is the backward difference method: ,in The time interval between the two frames is 1 / 30 of a second. Then, the sequence of horizontal shift velocity of the center of gravity during the entire standing process is paired with the force line imbalance intensity sequence obtained in step S22. A univariate linear regression analysis is performed on these two sets of data using the least squares method to find the best-fit line. Here, Let be the slope of the regression line. This is the intercept. The slope... This is defined as the centroid shift regression trend. For example, if the calculation yields... =0.05s, which physically means that for every 1m / s increase in the patient's center of gravity velocity on the horizontal plane, the expected increase in the force line imbalance of their back muscles is 0.05m. This slope It is output as a key parameter for subsequent robot-assisted decision-making.

[0050] Step S24: Based on the center of gravity offset regression trend, combined with the spatial coordinates of the posture trajectory and the posture-related muscle tension change data, generate robot-assisted output parameters.

[0051] In this embodiment of the invention, after obtaining the center of gravity shift regression trend, it needs to be converted into auxiliary output parameters for the nursing robot to perform actions. This process combines the spatial coordinates of the posture trajectory with posture-tension data. First, the direction and magnitude of the shift are determined based on the regression trend, for example, a lateral shift of 0.025m and a longitudinal displacement of 0.285m. Second, the magnitude of the support force that the robot should apply is determined based on the muscle tension margin. In the experiment, when the scapular muscle group imbalance leads to insufficient load, the robot arm needs to provide an additional 9N of support force. The support force calculation process uses the difference between the regression trend shift and the electromyographic tension data to obtain the missing compensation value. Finally, the compensation force is used as a parameter and output synchronously with the motion trajectory of the robot arm end effector. For example, from standing up to 4.20s, the auxiliary output parameters include: a support force of 9N from the right arm, a supporting force of 6N from the left arm, and a hand contact angle adjustment of 3°. These parameters are sent to the control terminal in real time to ensure that the robot can provide continuous and stable assistance when the patient's center of gravity rises, shifts laterally, and has muscle imbalance.

[0052] In another embodiment, an auxiliary triggering condition is set: when the intensity of force line imbalance detected in real time... When the distance exceeds a preset safety threshold of 0.03 meters, the system immediately generates auxiliary parameters. These parameters consist of three parts: the point of application, direction, and magnitude of the auxiliary force. The point of application is set to the current contact point between the robot arm's end effector and the patient's body; the three-dimensional coordinates of this point are calculated in real-time using the robot's forward kinematics. The direction of the auxiliary force is designed to be opposite to the direction of the force line imbalance to provide correction. Specifically, in the local two-dimensional coordinate system of the robot's hand, the direction vector of the force line imbalance is... Therefore, the direction vector of the auxiliary force is This two-dimensional vector is transformed into a three-dimensional force direction vector in the world coordinate system through the robot's end effector's attitude matrix. The magnitude of the auxiliary force... The calculation is performed using a proportional controller, the magnitude of which is proportional to the amount by which the force line imbalance exceeds a threshold, and is adjusted in conjunction with the trend of the center of gravity shift. Here, proportional gain It is a preset constant, such as 200 N / m. For example, when the real-time force line imbalance intensity is 0.035 m, the horizontal velocity of the center of gravity is 0.2 m / s, and the regression trend... When the speed is 0.05 seconds per meter, the calculated auxiliary force is: (N). Finally, this set of robot-aided output parameters, including the coordinates of the three-dimensional point of application, the three-dimensional force direction vector, and the force magnitude scalar, is packaged and sent to the robot control system.

[0053] Step S22 includes the following steps:

[0054] Step S221: Calculate the mean difference of tension intensity changes during posture changes based on posture-related muscle tension change data;

[0055] Step S222: Based on the average difference of tension intensity changes, perform three-dimensional force line direction vector decomposition on the posture-related muscle tension change data during posture changes to obtain a spatial force line distribution vector set;

[0056] Step S223: Perform orthogonal second-order statistical decomposition on the spatial force line distribution vector set to obtain the second-order decomposition data of the force lines;

[0057] Step S224: Analyze the force line imbalance intensity of muscle groups during posture changes by using the second-order decomposition data of force lines to analyze the spatial force line distribution vector set, thereby obtaining the force line imbalance intensity.

[0058] As an example of the present invention, reference is made to... Figure 3 As shown, step S22 in this example includes:

[0059] Step S221: Calculate the mean difference of tension intensity changes during posture changes based on posture-related muscle tension change data;

[0060] In this embodiment of the invention, the collected posture-related muscle tension change data are centrally organized, and the electromyographic signal amplitudes of the forearm, upper arm, and scapular back at each time point are uniformly expressed with their corresponding coordinate information. For example, in the experiment, the sampling period is set to 1ms, the total sampling time is 6.5s, and a total of 6500 data points are obtained. The electromyographic signal voltage amplitude is filtered and mapped to the muscle equivalent force value. The mapping relationship is established by the experimental calibration curve, such as a potential amplitude of 0.50V corresponding to a tension of 10N, and a potential amplitude of 0.70V corresponding to a tension of 14N. Subsequently, the tension values ​​of multiple contact areas within the same time series are weighted and averaged to obtain the comprehensive tension intensity of a single frame. Then, the average deviation is calculated by the difference between consecutive frames to obtain the average difference of tension intensity changes. Taking real experimental data as an example, within the time interval of 3.20s to 3.25s, the forearm tension increased from 10.2N to 11.6N, the back tension increased from 14.0N to 15.5N, and the scapular tension increased from 12.2N to 12.8N. The summation and averaging of these three changes yielded a mean difference of +1.7N for this time period. By continuously calculating the mean difference across all sampling intervals, a complete sequence of mean differences in tension intensity changes was obtained. This sequence reflects the overall fluctuation of force changes in different muscle groups during the standing-up process, providing a basis for subsequent three-dimensional vector decomposition.

[0061] In another embodiment, based on posture-related muscle tension change data, the short-term fluctuation of muscle tension intensity during posture changes is calculated and quantified as the mean difference of tension intensity changes. This calculation is performed within a sliding time window of 6 consecutive time points (i.e., 0.2 seconds). For each time point, the system first extracts the tension intensity data of three core muscle groups from the posture-related muscle tension change data. These three muscle groups correspond to the "back and scapular region," "hand / forearm region," and "elbow / upper arm region" collected in step S12, respectively. The tension intensity value of each region is obtained by arithmetic averaging of the root mean square voltage values ​​of its corresponding multiple electromyographic signal channels. For example, the tension intensity value of the "back and scapular region" is... It is obtained by adding the root mean square voltage values ​​of the corresponding four channels and dividing by 4. Then, within a sliding window, the first-order backward difference of these six consecutive tension intensity values ​​is calculated, resulting in five consecutive tension intensity variation values: ,in Iterate through the window from the second point to the sixth point. Finally, calculate the arithmetic mean of the absolute values ​​of these five changes; this result is the average difference in tension intensity change at the current time point. The calculation process is as follows: This sliding window calculation is repeated for every time point in the entire rising process, thereby generating a time series of tension intensity changes that is synchronized with each muscle group and characterizes the intensity of its tension fluctuations.

[0062] Step S222: Based on the average difference of tension intensity changes, perform three-dimensional force line direction vector decomposition on the posture-related muscle tension change data during posture changes to obtain a spatial force line distribution vector set;

[0063] In this embodiment of the invention, after calculating the average difference in tension intensity, it needs to be projected into three-dimensional space for directional decomposition. First, a coordinate reference system is defined, with the z-axis as the vertical direction, the y-axis as the front-back direction, and the x-axis as the left-right direction. Then, the average difference in tension intensity for each muscle group is projected into this three-dimensional coordinate system. Taking the time 3.40s in the experiment as an example, the average difference in tension intensity of the left forearm is +1.4N. Calculations based on joint posture trajectory show that the main force direction of the muscle makes an angle of 20° with the x-axis, 45° with the y-axis, and 55° with the z-axis. The resulting three directional components are: x-direction 0.48N, y-direction 0.99N, and z-direction 0.80N. The same method is applied to the back and scapular regions, yielding an average difference in tension intensity of -0.35N for the x-component, 1.20N for the y-component, and 1.55N for the z-component at the same time. In this way, the average difference in tension intensity of all muscle groups is decomposed into a vector form, forming a spatial force line distribution vector. The entire process of getting up is continuously calculated at each time point, generating a set of spatial force line distribution vectors, which is a set of multidimensional vector data that varies with time. This vector set clearly describes the direction of force contribution and its trend of change with the movement, and serves as the mathematical input set for further orthogonal decomposition.

[0064] Step S223: Perform orthogonal second-order statistical decomposition on the spatial force line distribution vector set to obtain the second-order decomposition data of the force lines;

[0065] In this embodiment of the invention, after obtaining the spatial force line distribution vector set, it is then decomposed using an orthogonal second-order statistical method. The purpose of the decomposition is to eliminate the linear coupling relationship between each direction and extract the independent principal variables. Specifically, the mean and variance of the three directional components are calculated for the vector set in each time period, and then the covariance matrix between different directions is obtained. Taking experimental data as an example, during the period from 3.60s to 3.70s, the mean in the x-direction is 0.52N, and the variance is 0.04(N²); the mean in the y-direction is 1.10N, and the variance is 0.06(N²); and the mean in the z-direction is 1.20N, and the variance is 0.05(N²). In the covariance matrix, the covariance in the xy-direction is 0.015, the covariance in the yz-direction is 0.018, and the covariance in the xz-direction is 0.012. Through orthogonal decomposition, this matrix is ​​transformed into three independent principal components, called the second-order decomposition vectors. The decomposition results show that within this time period, the first principal component, along the yz plane, contributes 62%, corresponding to a change intensity of 1.45 N; the second principal component, along the direction where the x-axis and z-axis make an angle of 40°, contributes 25%, corresponding to an intensity of 0.85 N; and the third principal component contributes 13%, corresponding to an intensity of 0.36 N. Through time-period decomposition, a second-order decomposition dataset of force lines was formed.

[0066] Step S224: Analyze the force line imbalance intensity of muscle groups during posture changes by using the second-order decomposition data of force lines to analyze the spatial force line distribution vector set, thereby obtaining the force line imbalance intensity.

[0067] In this embodiment of the invention, after obtaining the second-order decomposition data of force lines, it is used to analyze the force line imbalance intensity of different muscle groups during posture changes. The specific implementation process is as follows: within each time period, the force lines of the left and right symmetrical muscle groups and the anterior and posterior muscle groups are compared, and the average of the squared differences is used as the quantitative value of the imbalance degree. Taking 4.00s as an example in the experiment, the intensity of the second-order principal component formed by the left arm in the forward direction is 1.20N, and the intensity of the corresponding principal component of the right arm is 1.75N, with a difference of 0.55N; the intensities of the back muscles and scapular muscles in the vertical direction are 1.45N and 2.10N respectively, with a difference of 0.65N. By standardizing all differences, the imbalance intensity index is defined as the square root of the sum of the squares of all differences. This calculation at 4.00s yields a result of 0.85N. After normalizing this result with body mass-related parameters, the imbalance intensity ratio is found to be 32%. By calculating the imbalance intensity at every moment of the entire action process, a complete imbalance intensity can be generated. For example, the maximum value occurs in the interval between 3.80s and 4.20s, and the maximum imbalance intensity is 38%.

[0068] Step S23 includes the following steps:

[0069] Step S231: Deconstruct the spatial coordinates of the posture trajectory to obtain the rate of change of the patient's speed, joint range of motion, and trunk posture angle when the patient gets up during the nursing process; wherein the joints include the hip joint, knee joint, ankle joint, shoulder joint, and elbow joint; wherein the trunk refers to the area from the pelvis to the shoulder.

[0070] Step S232: Obtain the dynamic center of gravity trajectory based on the velocity change rate, joint movement angle, and trunk posture angle; perform polynomial fitting on the dynamic center of gravity trajectory to generate the center of gravity displacement fitting vector.

[0071] Step S233: Perform coordinate mapping processing on the force line imbalance intensity based on the attitude trajectory spatial coordinates to generate local force line imbalance distribution data corresponding to each spatial coordinate point;

[0072] Step S234: Based on the center of gravity displacement fitting vector, perform center of gravity trend vector difference on the local force line imbalance distribution data to obtain the center of gravity offset trend sequence;

[0073] Step S235: Perform correlation regression analysis on the center of gravity shift trend sequence, and then perform center of gravity shift regression deconstruction to obtain the center of gravity shift regression trend.

[0074] In this embodiment of the invention, the collected posture trajectory spatial coordinates are deconstructed frame by frame. A 3D camera outputs point cloud data of the patient's ascent process at a frame rate of 30Hz. The joint recognition module extracts the spatial coordinates of the hip, knee, ankle, shoulder, elbow, and lumbar spine center points. To obtain the rate of change of velocity, a difference is made between consecutive frames. For example, at 2.00s, the coordinates of the left knee joint are (0.20m, 0.32m, 0.51m), and at 2.03s, the coordinates are (0.20m, 0.325m, 0.53m). Dividing the difference by the time interval 0.03s yields the rate of change of velocity of the left knee joint as (0.0m / s, 0.17m / s, 0.67m / s). In joint angle calculation, the included angle between the two limb vectors connected by the joint is calculated; for example, the hip joint is formed by the thigh vector and the trunk vector. In the experiment, the left hip joint angle was 102° at 3.00s, gradually extending to 174° at 3.80s. Simultaneously, the trunk posture angle was calculated using the angle between the line connecting the pelvic point and the midpoint of the shoulder and the vertical axis. In the test case, the trunk forward tilt angle was approximately 25° at 2.50s, decreasing to 8° at 4.20s. These calculations yielded a complete time-series data matrix of velocity change rate, joint range of motion, and trunk posture angle, clearly describing the body's dynamic characteristics.

[0075] After deconstructing the joints and torso, the dynamic center of gravity trajectory during the standing process needs to be calculated. The center of gravity position is obtained by weighted averaging of the hip, knee, ankle, shoulder, and midpoint of the torso, with weighting coefficients set according to the proportion of each part in the total body mass. For example, the weight of the two hip joints and the lumbar spine center point is 0.40, the weight of the two knee joints is 0.25, the weight of the two shoulder joints is 0.20, and the weight of the two ankle joints is 0.15. During the experiment, the patient's initial center of gravity position was (0.215m, 0.175m, 0.835m), gradually shifting with the movement, reaching (0.225m, 0.240m, 1.120m) at 4.80s, with a longitudinal displacement of approximately 0.285m and a lateral displacement of approximately 0.010m. The collected center of gravity trajectory was interpolated to form a continuous sequence over time. Subsequently, a third-order polynomial fitting was performed on the center of gravity trajectory, and the fitting expression is: ,in Indicates the time taken to stand up, in seconds (s). The initial centroid position component, The coefficient of the center of gravity velocity term reflects a linear change. The coefficient of the acceleration term at the center of gravity. These are high-order nonlinear coefficients. Taking the longitudinal component of the experimental centroid as an example, the coefficient of determination of the fitting result is 0.98, indicating that the fitted curve can accurately reflect the true trajectory. Finally, the centroid displacement fitting vector is obtained, and the vector elements include continuous fitting terms in the x, y, and z directions, laying a mathematical foundation for subsequent trend analysis.

[0076] After obtaining the dynamic center of gravity fitted trajectory, it is necessary to establish a mapping relationship between the force line imbalance intensity data and the attitude trajectory spatial coordinates. First, the force line imbalance intensity comes from the calculation result of step S22 and varies with time in units of N. Second, based on the position of the attitude trajectory in three-dimensional coordinates, the local spatial point at each moment is determined. For example, at 3.60s, the center of gravity coordinates are (0.220m, 0.205m, 0.980m), and the force line imbalance intensity at this time is 0.82N. The two are bound by a timestamp, and this intensity is assigned to the corresponding center of gravity position. This operation is performed for all moments, forming a mapping set of "coordinate point - imbalance intensity". To further obtain the local distribution, joint coordinates are also included in the mapping range. For example, at 3.70s, the coordinates of the right knee joint (0.185m, 0.315m, 0.540m) were assigned a 0.65N imbalance intensity component, and the coordinates of the left shoulder joint (0.245m, 0.210m, 1.200m) were assigned a 0.95N value. After processing each joint and center of mass, local imbalance distribution data covering key points throughout the body was generated. This data not only includes spatial location but also describes the current state of force imbalance, revealing the correspondence between the center of gravity trend and the force imbalance of muscle groups in subsequent regression analysis, providing the necessary accurate data for further generation of robot output parameters.

[0077] The center of gravity displacement fitting vector obtained in step S232 is used. This vector is a three-dimensional function that varies with time, reflecting the displacement trends in the x, y, and z directions, respectively. Next, using the local force line imbalance distribution data generated in step S233, the imbalance intensity value of each joint and center of gravity position is synchronized with the corresponding spatial coordinates. To calculate the center of gravity trend difference, the correspondence between the center of gravity fitting vector and the imbalance intensity at that position is first determined at the same time point. For example, at 3.80s, the center of gravity coordinates are (0.222m, 0.218m, 1.015m), the fitting vector outputs a longitudinal displacement velocity of 0.25m / s, and the corresponding local imbalance intensity is 0.82N. Using this moment as a reference, the data of adjacent frames are differentiated to obtain the difference vector under the combined effect of center of gravity displacement change and local imbalance intensity. Specifically, the difference between the fitting vector at time points t and t+Δt is calculated, and then combined with the imbalance value to form a correction vector. For example, in the interval from 3.80s to 3.83s, the longitudinal displacement increases by 0.012m and the lateral displacement increases by 0.002m, corresponding to an increase of 0.20N in the imbalance correction component. This results in a differential trend in this interval of "rising with enhanced left-side offset." By repeating the differential operation on the entire time series, a center of gravity offset trend sequence is finally formed. This sequence, with time as the main axis, records the displacement direction and rate offset trend of the center of gravity in three-dimensional space and is coupled with the imbalance intensity, accurately reflecting the dynamic characteristics of the center of gravity at each stage during the patient's ascent.

[0078] After generating the trend sequence, regression analysis is required to extract the overall pattern. Specifically, the center of gravity shift trend sequence is divided into three-dimensional components: a horizontal component x(t), a vertical component y(t), and a vertical component z(t). For each component, regression is performed using the correlation between the time series and the intensity of the imbalance. Taking the vertical component as an example, its function over time is defined as follows: ,in, Indicates the longitudinal center of gravity shift trend. Indicates time, Indicates the imbalance intensity sequence at time... The numerical value is in N. This is the initial vertical offset position. The longitudinal velocity regression coefficient, For acceleration regression coefficients, The coefficient represents the imbalance coupling coefficient. Taking experimental data as an example, during the period from 2.50s to 4.50s, the total longitudinal upward displacement was approximately 0.285m, with a maximum imbalance intensity of 0.95N, corresponding to a coefficient of determination of 0.96 in the longitudinal regression expression. The results for lateral offset show a maximum offset of 0.025m, and regression analysis indicates that the imbalance term accounts for 38% of the trend contribution. The deconstruction process obtains a clear offset trend structure by splitting the regression function into a velocity trend component, an acceleration trend component, and an imbalance coupling component. For example, at 4.00s, the longitudinal trend is dominated by the velocity term, while the lateral trend is dominated by the imbalance component, indicating that the center of gravity rises steadily but the lateral offset is significant during this period. The final obtained center of gravity offset regression trend is a multidimensional time function containing three components, used to guide the generation of auxiliary output parameters for the nursing robot.

[0079] Performing centroid trend vector differencing on the aforementioned local force line imbalance distribution data includes the following steps:

[0080] Based on the aforementioned center of gravity displacement fitting vector and local force line imbalance distribution data, spatiotemporal coordinate cross-matching is performed to obtain the center of gravity-force line coupling mapping data corresponding to each time-series coordinate point.

[0081] Based on the aforementioned centroid-force line coupling mapping data, differential vector fields are constructed for adjacent time-series coordinate points to obtain a local non-uniformity differential vector field for continuous time series.

[0082] The local non-uniformity differential vector field is subjected to gradient difference processing at each spatial coordinate point to obtain a local centroid offset gradient sequence.

[0083] Multi-scale trend coupling is performed on the local centroid shift gradient sequence, and the shift change rate at different time scales is combined to obtain the multi-scale centroid shift trend.

[0084] The centroid shift trend sequence is obtained by performing centroid shift trend vector difference based on multi-scale centroid shift trend.

[0085] In this embodiment of the invention, the centroid displacement fitting vector obtained in step S232 is called, which provides the centroid position coordinates at each time point. And the displacement trends in three directions. Meanwhile, the local force line imbalance distribution data provided in step S233 includes the position of each joint. The force line imbalance intensity value F is used. To achieve cross-matching, these two types of data need to be aligned at the same time stamp. For example, at 3.80s, the center of gravity fitted vector coordinates are (0.222m, 0.218m, 1.015m), and the longitudinal velocity is 0.25m / s; at this time, the right knee joint coordinates are (0.185m, 0.315m, 0.540m), with an imbalance intensity value of 0.68N, and the left shoulder joint coordinates are (0.245m, 0.210m, 1.200m), with an imbalance intensity value of 0.92N. Through spatiotemporal coordinate cross-matching, the corresponding center of gravity vector is mapped to the local imbalance point of the joint. Specifically, the difference between the center of gravity coordinates and the joint coordinates is calculated as a relative position vector, and then the imbalance intensity is superimposed on it. For example, the displacement vector of the right knee joint relative to the center of gravity is (-0.037m, 0.097m, -0.475m), with an associated force line imbalance intensity of 0.68N, ultimately generating coupled data points (−0.037, 0.097, −0.475; 0.68). This operation is performed on all joints and center of gravity positions to obtain a complete “center of gravity-force line” coupled mapping dataset for each time step. It contains both spatial positional relationships and mechanical inhomogeneity characteristics, serving as the fundamental input for subsequent differential vector field construction.

[0086] After obtaining the spatiotemporal cross-mapping data, a continuous temporal differential vector field needs to be constructed. First, differential operations are performed on the coupled mapping data at every two adjacent time points. The differential process includes simultaneous calculation of positional changes and imbalance intensity changes. For example, the center-of-gravity-right knee coupling data at 3.80s and 3.83s are (-0.037, 0.097, -0.475; 0.68) and (-0.040, 0.102, -0.462; 0.72), respectively. Subtracting the two yields the differential vector (-0.003m, 0.005m, 0.013m; 0.04N). Here, the first three results are temporal displacement differences, and the last term is the imbalance intensity difference. In this way, the differential values ​​of all joints relative to the center of gravity are calculated. Next, these differential values ​​are concatenated to form a complete vector field. Each vector element in this field represents both the trend of displacement direction changes and the increase or decrease in mechanical inhomogeneity. For example, at 3.83s, the difference results for the left shoulder joint are (0.003m, 0.004m, 0.015m; 0.06N), indicating that the spatial movement of the shoulder relative to the center of gravity during this stage is forward and upward, accompanied by an increase in imbalance. Combining the difference vectors of all joints and the center of gravity forms a continuous temporal local inhomogeneity difference vector field. This vector field can continuously describe the coupling effect of the patient's center of gravity shift trend and local mechanical anomalies during the standing process in the time dimension, providing accurate input data for subsequent trend analysis and regression deconstruction.

[0087] First, the continuous temporal local inhomogeneity difference vector field generated in the previous steps is invoked. This vector field contains the relative displacements and directions of multiple joints relative to the center of mass, as well as the corresponding force line imbalance difference values. At each time stamp, the vector field uses three-dimensional coordinate points as nodes. For example, at the end of 3.80s, the difference result for the right knee node is (-0.003m, 0.005m, 0.013m; 0.04N), and the difference result for the left shoulder node is (0.003m, 0.004m, 0.015m; 0.06N). To obtain local spatial gradient features, gradient calculations need to be performed on the difference vectors of adjacent joint nodes in the three-dimensional coordinate domain. Specifically, at each joint node, the difference between its three-dimensional difference vector and the vectors of its surrounding adjacent nodes is calculated, and then divided by their spatial distance. For example, if the difference vector difference between the shoulder and elbow joints, which are 0.12m apart, is (0.006m, -0.002m, 0.010m), then the local gradient is (0.050, -0.017, 0.083), in dimensionless ratio. This method is used to perform similar calculations on all joints, forming a complete local gradient matrix at each time stamp. This matrix reflects the non-uniform vector change rate at different locations. Arranging the matrices from all time stamps consecutively yields the local center of gravity offset gradient sequence. This sequence clearly expresses the offset gradient trajectory of each local point during center of gravity movement, laying the foundation for subsequent cross-scale coupling.

[0088] The local centroid shift gradient sequence only reflects data at a single time resolution. To reflect the combined patterns of short-term jerking and long-term shift throughout the patient's ascent, multi-scale processing is required. The specific steps are as follows: The time series is divided into three scales: a fine scale of 0.03s (corresponding to a 30Hz camera sampling rate); a medium scale of 0.30s (corresponding to a window of 10 frames); and a macro scale of 1.50s (corresponding to a window of 50 frames). At each scale, the mean and rate of change of the gradient sequence are calculated. For example, within the 0.30s window, when the local gradient mean is (0.045, 0.020, 0.078) and the rate of change is (0.003, 0.002, 0.006), it indicates a significant upward trend in the centroid during that period. Subsequently, the data from different scales are synthesized according to weights, with a weighting principle of 0.4 for the fine scale, 0.35 for the medium scale, and 0.25 for the macro scale, to ensure that small-amplitude, short-term disturbances do not mask the overall trend. Based on experimental data, during the 3.00s to 4.50s interval when the patient stood up, fine-scale data showed multiple local lateral shifts of up to 0.008m, while the macro-scale trend showed a continuous vertical increase of 0.28m. The fused results formed a multi-scale center of gravity shift trend, which not only included the overall displacement direction but also accurately captured the superposition effect at different time scales, thus better reflecting the actual dynamic process.

[0089] After obtaining the multi-scale centroid shift trend, further processing yields the final centroid shift trend sequence. The specific process is as follows: First, three-directional component curves are extracted from the multi-scale trend. For example, the vertical trend increases from an initial 0.835m to 1.120m, the horizontal trend fluctuates between 0.005m and 0.025m, and the overall shift in the forward and backward directions (y-axis) does not exceed 0.010m. Then, these trend curves are progressively subtracted to obtain the trend vector difference. For example, between 3.50s and 3.53s, the vertical difference value is (+0.012m), and the horizontal difference value is (+0.002m). The vertical difference corresponds to a gradient increment of +0.05, and the horizontal difference corresponds to an increment of +0.08. The difference results reflect the instantaneous change strength of the centroid trend. The difference results from all time periods are concatenated to form a complete centroid shift trend sequence. This sequence is represented as a set of three-dimensional time functions, recording the gain direction and magnitude of the centroid trend at each moment. For example, in the experimental data, the maximum longitudinal trend increment occurred in the interval from 3.95s to 3.98s, with a value of +0.018m, while the maximum lateral trend increment occurred in the interval from 4.05s to 4.08s, with a value of +0.004m. Through this processing, a complete sequence of center of gravity shift trends over time can be output, providing input data for the nursing robot to further generate precise mechanical auxiliary parameters.

[0090] Step S24 includes the following steps:

[0091] Step S241: Based on the center of gravity shift regression trend, combined with the spatial coordinates of the posture trajectory, determine the patient's postural tendency direction when getting up;

[0092] Step S242: Perform fatigue load analysis on posture-related muscle tension change data to determine the acceptable force margin for muscles;

[0093] Step S243: Optimize multi-point contact mechanics based on the orientation of the posture and the acceptable force margin of the muscles to obtain dynamic contact force;

[0094] Step S244: Perform adaptive fuzzy inverse kinematics transformation of the robot based on dynamic contact force and orientation to obtain the robot's joint space motion trajectory;

[0095] Step S245: Based on the robot joint spatial motion trajectory and dynamic contact force, perform data parameter fusion to generate robot auxiliary output parameters.

[0096] In this embodiment of the invention, the center of gravity offset regression trend sequence obtained in step S235 is invoked. This sequence includes the dynamic displacement trends in the x, y, and z coordinate directions. For example, in the experimental record, the average displacement in the longitudinal direction (z) is 0.285m, the maximum displacement in the lateral direction (x) is 0.025m, and the offset in the front-to-back direction (y) does not exceed 0.010m. Simultaneously, the posture trajectory spatial coordinate sequence provides the angles of key joints. For example, during the standing process, the left hip joint angle increases from 102° to 172°, the right knee joint angle extends from 85° to 176°, and the trunk tilt angle gradually decreases from 25° to 8°. By synchronizing the offset trend with the joint angles, the overall orientation of the posture tendency can be determined. For example, in the period from 3.80s to 4.10s, the longitudinal trajectory shows an upright trend, while the lateral peak offset reaches 0.022m, and the shoulder is offset to the left by approximately 6° relative to the center of gravity. The overall posture tendency orientation is determined to be "vertical ascent, leftward offset". The final posture tendency vector consists of two parts: the center of gravity direction [0.022m, 0.008m, 0.285m] and the posture angle [θhip=172°, θknee=176°, θtorso=8°]. Using the posture-related muscle tension change data obtained in step S1, fatigue load analysis was performed. Specifically, the cumulative tension integral of the muscle groups over different time periods was calculated and compared with the physiological limit. During the experiment, the peak electrical signal voltage of the forearm muscles reached 0.65V during the rising phase (total duration 6.5s), corresponding to an equivalent tension of approximately 12.8N. The tension was cumulatively integrated over time, yielding a total stress value of 58N·s. The standard physiological endurance threshold is 72N·s, therefore the remaining usable margin is 14N·s. The analysis results for the back muscles showed a stress value of 130N·s and a margin of 40N·s; the corresponding margin for the scapular muscles was 18N·s. By comparing the three sets of data, it can be seen that the back muscles have the largest load margin, while the scapular muscles have the smallest. Therefore, it is determined that in robot-assisted control, the shoulder support force should not exceed its margin of 18N, the forearm should not exceed 14N, and the allowable compensation range for the back is within 40N. This conclusion provides a strict upper limit for force in multi-point contact optimization, ensuring that contact force control is scientific and does not exceed the patient's physical capabilities.

[0097] The postural tendency orientation of S241 is combined with the acceptable force margins for different muscle groups in S242 for calculation. First, the contact points are defined as three main areas: the left forearm contact area, the right forearm contact area, and the back support area. At 3.90s, the postural tendency is determined to be a 0.022m leftward shift and a 0.285m longitudinal ascent. At this point, the lateral damping force borne by the left forearm contact should be higher than that of the right to correct the shift. Based on the margin, the maximum allowable force for the left arm is 14N, therefore an actual force of 9N is allocated; the right arm is allocated 6N to maintain balance. The back, as the longitudinal support area, bears a longitudinal compensation force of 12N, lower than the maximum margin of 40N, ensuring safety. The optimization process uses a matrix balancing method, coupling the force output of the three contact areas with the center of gravity orientation to ensure that the resultant force direction aligns with the main longitudinal upward direction while simultaneously counteracting the lateral shift. The final output is a dynamic contact force vector set: left forearm (9N, -0.002m), right forearm (6N, 0.003m), and back (12N, 0.000m). These parameters ensure overall balance and serve as precise force inputs for the robot's actuators.

[0098] Based on the dynamic contact force vector and orientation obtained from S243, the robot joint motion trajectory is constructed. The nursing robot arm has a 6-DOF joint structure, namely the shoulder joint (3 rotation axes), elbow joint (1 rotation axis), and wrist joint (2 rotation axes). The input parameters are the contact forces Fx, Fy, and Fz at the point of application and the target orientation angles θtorso, θhip, and θknee. Using inverse kinematics, the target position of the end effector at time t is mapped to a sequence of joint angles. For example, at 3.95s, the back support force of 12N corresponds to vertical support upward movement, requiring the end effector to rise by 0.012m. The calculated result is a shoulder joint rotation angle of +0.06rad and an elbow joint extension angle of +0.04rad. Simultaneously, the leftward offset of the orientation corresponds to the end effector movement direction (0.004m, -0.002m), and the calculated wrist joint rotation angle around the y-axis is -0.03rad. A fuzzy control strategy is used to ensure that the input dynamic contact force can still solve for an effective trajectory within a margin range. That is, when the target and the solved joint angles are not perfectly matched, the longitudinal support requirement is prioritized to minimize the error. This calculation is executed continuously, generating a smooth sequence of joint space trajectories over the entire 6.5s range. The joint angle update frequency is maintained at 50Hz, and the output is the joint angle value and angular velocity at each moment.

[0099] Two types of data are fused into auxiliary output parameters. One type is the joint motion trajectory data obtained from S244, which consists of six joint angle values ​​and angular velocities at each time point. For example, at 4.00s, the shoulder joint angle is (0.80rad, 0.20rad, 0.35rad), the elbow joint extension is 0.45rad, and the wrist joint angle is (-0.15rad, 0.10rad). The other type is the dynamic contact force vector obtained from S243, such as 9N for the left arm, 6N for the right arm, and 12N for the back at the same time point. The data fusion method is as follows: at each time point, the joint trajectory sequence is coupled one-to-one with the contact force vector group to generate an extended parameter matrix. For example, the matrix unit form is [joint angle, joint velocity, contact force magnitude, contact offset direction]. Taking 4.00s as an example, the left arm joint combination angle of 0.75rad corresponds to a contact force of 9N, and the coupling forms the unit {0.75, 0.03rad / s, 9, -0.002m}. All joint data and contact forces are combined and output to form a parameter matrix of six degrees of freedom × contact points. The resulting auxiliary output parameter sequence contains a time axis from 0s to 6.5s, achieving a comprehensive expression of motion and force. This parameter package is directly sent to the nursing robot's execution controller, driving the robotic arm to apply precise auxiliary operations that match the patient's real-time dynamic state during the patient's ascent.

[0100] Step S3 includes the following steps:

[0101] Step S31: Perform parameter feature importance learning on the robot's auxiliary output parameters to obtain auxiliary learning parameters;

[0102] Step S32: Encode instructions based on auxiliary learning parameters to obtain auxiliary parameter instructions;

[0103] Step S33: Using auxiliary parameter instructions as input, design the control architecture of the nursing robot to build a robot-assisted nursing architecture;

[0104] Step S34: Send the robot-assisted nursing architecture to the control terminal of the nursing robot to execute the control of the nursing robot.

[0105] In this embodiment of the invention, the robot-aided output parameter sequence generated in step S24 contains data in multiple dimensions, such as joint angles, angular velocities, contact force vectors, and contact direction correction values. For importance analysis, the parameters need to be expanded along the time dimension. For example, at 3.80s, the left arm contact force is 9N, the joint angle is 0.75rad, the joint angular velocity is 0.03rad / s, and the back contact force is 12N. By statistically analyzing the parameters over the entire 6.5s time period, the contribution of each type of parameter to the overall trend is first calculated. The method is as follows: using the center of gravity shift trend as the target factor, the correlation values ​​between different parameters and the target are calculated. Experiments show that the correlation coefficient between the contact force vector and the center of gravity shift reaches 0.87, while the correlation coefficient between the joint velocity and the center of gravity shift is only 0.54. Further verification using variance contribution shows that the variance of the longitudinal contact force is 2.3 (N²), accounting for 41% of the overall index. Therefore, the contact force parameter has the highest importance weight, followed by the joint angle, and then the angular velocity. The final output of the auxiliary learning parameters is represented by a set of weight coefficients, such as contact force = 0.45, joint angle = 0.35, and angular velocity = 0.20. This set of parameters serves as the input for subsequent steps.

[0106] After completing the feature importance analysis, instructional codes need to be generated. The process is as follows: for each category parameter, encoding is performed according to the weights assigned in S31. For example, in the control of joint angle and angular velocity, weights of 0.35 and 0.20 are used to adjust them, merging the two types of inputs into an angle control factor. In contact force vector control, a weight of 0.45 is used to highlight its priority in the support process. The encoding rule uses an integer method, mapping each parameter to a fixed-width encoding bit. For example, the contact force parameter in the range of 0–30N is mapped to an integer value of 0–255, and the joint angular velocity in the range of 0–0.2rad / s is mapped to an integer value of 0–255. Taking a real experiment as an example, at 4.00s, the contact force of the left arm is 9N, encoded as 77, the contact force of the right arm is 6N, encoded as 51, and the contact force of the back is 12N, encoded as 102. The joint angular velocity of 0.03rad / s is encoded as 38, and the corresponding joint angle of 0.75rad is encoded as 136. The final combination forms a set of instructional data packets, such as [77,51,102,38,136]. This data packet is the auxiliary parameter instruction, representing the action posture and force distribution that the robot needs to perform at that moment.

[0107] Assistive parameter commands are used as inputs to the control architecture of the nursing robot. First, the command elements are decomposed, comprising three types of factors: functional factors, temporal factors, and constraint factors. Functional factors are the action commands for different joints, such as shoulder elevation, elbow extension, and wrist deflection; temporal factors specify the refresh frequency, set to 50Hz in this experiment; constraint factors specify the mechanical range, such as output force not exceeding 30N and angular velocity not exceeding 0.2rad / s. These factors are then combined to form a primary control matrix. For example, at 4.00s, the functional factor matrix is ​​{shoulder elevation 0.75rad, force output 9N; elbow extension 0.45rad, force output 6N; back support position 0.012m, force output 12N}. This matrix is ​​logically superimposed to form a control framework, which is further fused with the temporal factors to obtain an hourly updated assisted nursing architecture. Finally, priority rules are set in the architecture, such as longitudinal support commands taking precedence over lateral correction commands. When there is a conflict between the longitudinal direction and the horizontal deflection, the constraint factors ensure that longitudinal movements are given priority in control resource allocation. This design ensures stable execution of the robot architecture under complex movements.

[0108] In the final step, the generated nursing architecture needs to be sent to the robot control terminal and executed. The specific process is as follows: First, the nursing architecture generated by S33 is converted into communication data packets, including joint control sub-instructions and contact force control sub-instructions. Each data packet contains a timestamp (accuracy 1ms), target position parameters (in meters), target angle parameters (in rads), and target force parameters (in N). For example, the data packet content at 4.00s is: shoulder joint target angle 0.75 rad, force 9 N; elbow joint target angle 0.45 rad, force 6 N; back contact target displacement 0.012m, force 12 N. The data packets are transmitted to the robot control terminal via a bus at a frequency of 50Hz. After receiving the data, the terminal directly drives the actuators, including the motor controller, force sensor feedback module, and end effector. The control terminal internally samples the feedback signal at millisecond intervals and compares it in real time with the target values ​​given in the input nursing architecture, making minor corrections when necessary. The entire process ensures that the robotic arm performs continuous, stable, and precise auxiliary movements during the 6.5s of the patient getting up.

[0109] Step S33 includes the following steps:

[0110] Step S331: Take the auxiliary parameter instruction as input, perform instruction element decomposition processing on the auxiliary parameter instruction to obtain a multi-dimensional instruction factor set containing instruction function factor, timing factor and constraint factor;

[0111] Step S332: Based on the aforementioned multidimensional instruction factor set, deduce the functional interaction coupling relationship of the nursing robot to obtain the functional interaction coupling relationship;

[0112] Step S333: Learn the policy association of the functional interaction coupling relationship according to the policy gradient algorithm to obtain the interaction coupling association policy;

[0113] Step S334: Design the control architecture of the nursing robot based on the interaction coupling association strategy to build a robot-assisted nursing architecture.

[0114] In this embodiment of the invention, the input is the auxiliary parameter instruction obtained in step S32. This instruction is a combination of data encoded from the contact force vector, joint angle, angular velocity, and timestamp. For example, at 4.00s, the instruction format is [9N, 6N, 12N, 0.75rad, 0.45rad, 0.03rad / s, 4000ms]. To construct a multi-dimensional factor set, this composite instruction needs to be decomposed into three types of elements: functional factors, temporal factors, and constraint factors. The functional factors describe the specific task to be performed, such as lifting the shoulder joint, supporting the back, and lifting the forearm; the temporal factors specify the refresh frequency, which is 50Hz under the experimental conditions, corresponding to a period of 0.02s; the constraint factors specify physical limitations, such as single joint output force ≤30N, joint angular velocity ≤0.2rad / s, and maximum joint angle ≤2.0rad. The decomposition process maps the original encoded data to a three-dimensional vector to form a functional matrix (such as the joint target angle and action torque), a temporal matrix (such as the periodic sampling time interval), and a constraint matrix (such as the safety torque limit). Taking 4.00s as an example, the functional factors are defined as: shoulder joint angle 0.75 rad and force 9 N, elbow joint angle 0.45 rad and force 6 N, back translation 0.012 m and force 12 N; the timing factor is the refresh interval 0.02s; and the constraint factor is the upper limit of output force 30 N. This forms a complete set of multi-dimensional instruction factors.

[0115] The multidimensional factor set obtained from S331 is used as the basic input. During nursing care, multiple functional factors influence each other. For example, shoulder lifting and back support are mechanically coupled, and forearm flexion and elbow rotation are angularly linked. The method for deriving functional interaction relationships is as follows: First, a dependency matrix between factors is established. Taking shoulder lifting (9N) and back support (12N) as an example, if there is a 10° angle between their action directions, the coupling ratio of their mechanical components is cos(10°) = 0.985, indicating that the two forces are almost in the same direction and need to be coordinated synchronously. Furthermore, at the temporal level, if the shoulder joint target angle update is delayed by 0.01s at time 3.95s, it will cause a misalignment between the force output and the back support action, resulting in a temporal coupling coefficient of 0.5 in the dependency matrix. Finally, an interaction relationship matrix is ​​formed, where each unit value represents the coupling strength between factors, ranging from 0 to 1. For example: shoulder-back coupling strength 0.85, elbow-forearm coupling strength 0.78, shoulder-elbow coupling strength 0.62. This yields a functional interaction coupling relationship, which can clearly express the interaction law of different instruction factors under spatial mechanics and time constraints.

[0116] The functional interaction coupling relationship obtained in step S332 is further transformed into an operable strategy in the control execution process. First, the input needs to be defined, namely the interaction coupling matrix, which consists of the interaction strengths between multi-dimensional factors. For example, in the actual measurement process (standing up time 3.50s~4.20s), the coupling coefficient between shoulder joint elevation and back support is 0.85, the coupling coefficient between elbow joint extension and forearm lifting is 0.78, and the coupling coefficient between the shoulder joint and elbow joint is 0.62. Each coefficient represents the dependence strength of the corresponding functional factor in the action execution. In strategy association learning, a shift error function needs to be constructed, using the difference between the actual displacement of the center of gravity and the target fitted displacement as an optimization index to correct the control signal. The error function is defined as: ,in , , These represent the values ​​of the desired centroid displacement in the x, y, and z directions, respectively. , , This represents the actual displacement of the center of gravity; E(t) is the error value of the center of gravity at time t. This represents the time parameter. During the association learning process, for each coupling pair (i,j), the output weights are defined. The weights are increased or decreased using a gradient approach to ensure a decrease in the error function. For example, when the center of gravity shifts laterally by 0.014m at 3.90s, the coupling matrix indicates a relationship strength of 0.85 between the shoulder joint and back support, meaning the shoulder joint's motion compensation must be linked to the longitudinal force output of the back support. Through policy updates, the shoulder joint force is corrected from 9.0N to 9.5N, reducing the lateral shift to 0.005m. The policy learning process is executed frame-by-frame, with each update cycle lasting 0.02s, corresponding to a refresh rate of 50Hz. Experimental data shows that within the 3.50s~4.50s interval, the longitudinal center of gravity error decreases from 0.030m to 0.008m, and the lateral shift error decreases from 0.014m to 0.004m. Finally, after full data iteration, the output interactive coupling strategy consists of a set of priority weights: longitudinal support control weight 0.45, lateral correction control weight 0.35, and joint velocity correction weight 0.20.

[0117] The interactive coupling strategy obtained through S333 was used to design the control architecture as a hierarchical control system. The first layer is the center of gravity stabilization layer, which mainly handles longitudinal support movements, ensuring that the body's center of gravity moves upward synchronously with the body's ascent. The second layer is the offset correction layer, which allocates elbow and shoulder joint correction movements in real time based on lateral and forward / backward offset trends. The third layer is the constraint safety layer, which is responsible for monitoring whether all joint and force outputs are within safety boundaries, and immediately performs amplitude limiting if they exceed the preset upper limit. In the actual experiment, at 4.00s, the auxiliary control command was: the back actuator applied a 12N thrust, the shoulder joint applied a 9N lifting force, and the elbow joint maintained a 0.45rad angle. Because the coupling association strategy sets priorities, the back thrust signal is executed immediately through the first-level control channel, while the elbow joint signal is output after offset correction through the second-level channel. The timing control is executed continuously at a 50Hz refresh rate, ensuring that all joint outputs and auxiliary forces are updated once every 20ms. Finally, the robot-assisted nursing architecture was constructed as a set of hierarchical processing structures, including the functional factors, time series parameters, and constraint detection modules of each control channel, completely realizing the control mapping from strategy to execution.

[0118] The present invention also provides a control system for a nursing robot, for executing the control method for the nursing robot described above, the control system comprising:

[0119] The data acquisition module is used to acquire the patient's posture trajectory when getting up during the nursing process using a 3D camera; and to acquire the muscle tension change signals in the body contact area when the patient gets up during the nursing process using an electromyography (EMG) sensor based on the posture trajectory; and to perform synchronous correlation processing on the muscle tension change signals based on the posture trajectory to obtain posture-related muscle tension change data.

[0120] The robot output parameter analysis module is used to analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data. Subsequently, it performs center of gravity offset regression deconstruction to obtain the center of gravity offset regression trend; and generates robot auxiliary output parameters based on the center of gravity offset regression trend.

[0121] The control architecture design module is used to design the control architecture of the nursing robot by taking the robot-assisted output parameters as input, so as to build the robot-assisted nursing architecture and send it to the control terminal of the nursing robot to execute the control of the nursing robot.

[0122] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A control method for a nursing robot, characterized in that, An electromyography (EMG) sensor is deployed on the palm of the robotic arm of a nursing robot, and a 3D camera is deployed in the head region of the nursing robot. The control method of the nursing robot includes the following steps: Step S1: Acquire the patient's posture trajectory when standing up during the nursing process using a 3D camera; acquire muscle tension change signals in the body contact area when the patient stands up using an electromyography (EMG) sensor based on the posture trajectory; perform synchronous correlation processing on the muscle tension change signals based on the posture trajectory to obtain posture-related muscle tension change data. Step S2: Analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data. Then, perform center of gravity offset regression deconstruction to obtain the center of gravity offset regression trend. Generate robot-aided output parameters based on the center of gravity offset regression trend. Step S2 includes the following steps: Step S21: Construct the spatial coordinates of the posture trajectory during the standing process to obtain the spatial coordinates of the posture trajectory; Step S22: Analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data to obtain the force line imbalance intensity. Step S23: Based on the attitude trajectory spatial coordinates, perform a center of gravity offset regression deconstruction on the force line imbalance intensity to obtain the center of gravity offset regression trend; Step S24: Based on the center of gravity offset regression trend, combined with the spatial coordinates of the posture trajectory and the posture-related muscle tension change data, generate robot-assisted output parameters; Step S22 includes the following steps: Step S221: Calculate the mean difference of tension intensity changes during posture changes based on posture-related muscle tension change data; Step S222: Based on the average difference of tension intensity changes, perform three-dimensional force line direction vector decomposition on the posture-related muscle tension change data during posture changes to obtain a spatial force line distribution vector set; Step S223: Perform orthogonal second-order statistical decomposition on the spatial force line distribution vector set to obtain the second-order decomposition data of the force lines; Step S224: Analyze the force line imbalance intensity of muscle groups during posture changes by using the second-order decomposition data of force lines to analyze the spatial force line distribution vector set, thereby obtaining the force line imbalance intensity. Step S23 includes the following steps: Step S231: Deconstruct the spatial coordinates of the posture trajectory to obtain the rate of change of the patient's speed, joint range of motion, and trunk posture angle when the patient gets up during the nursing process; wherein the joints include the hip joint, knee joint, ankle joint, shoulder joint, and elbow joint; wherein the trunk refers to the area from the pelvis to the shoulder. Step S232: Obtain the dynamic center of gravity trajectory based on the velocity change rate, joint movement angle, and trunk posture angle; perform polynomial fitting on the dynamic center of gravity trajectory to generate the center of gravity displacement fitting vector. Step S233: Perform coordinate mapping processing on the force line imbalance intensity based on the attitude trajectory spatial coordinates to generate local force line imbalance distribution data corresponding to each spatial coordinate point; Step S234: Based on the center of gravity displacement fitting vector, perform center of gravity trend vector difference on the local force line imbalance distribution data to obtain the center of gravity offset trend sequence; Step S235: Perform correlation regression analysis on the center of gravity shift trend sequence, and then perform center of gravity shift regression deconstruction to obtain the center of gravity shift regression trend; Step S3: Use the robot-assisted output parameters as input to design the control architecture of the nursing robot, so as to build a robot-assisted nursing architecture, and send it to the control terminal of the nursing robot to execute the control of the nursing robot.

2. The control method for the nursing robot according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Capture the patient's posture trajectory when getting up during the nursing process using a 3D camera; Step S12: Using an electromyography (EMG) sensor and based on the posture trajectory, collect muscle tension change signals in the body contact areas when the patient gets up during the nursing process; wherein the body contact areas include: hand / forearm area, back and scapula area and elbow / upper arm area; Step S13: Filter the muscle tension change signal for noise and then embed the timestamp to obtain the time series of the change signal; Step S14: Based on the posture trajectory, perform synchronous correlation processing on the time series of the change signal to obtain posture-related muscle tension change data.

3. The control method for the nursing robot according to claim 1, characterized in that, Step S24 includes the following steps: Step S241: Based on the center of gravity shift regression trend, combined with the spatial coordinates of the posture trajectory, determine the patient's postural tendency direction when getting up; Step S242: Perform fatigue load analysis on posture-related muscle tension change data to determine the acceptable force margin for muscles; Step S243: Optimize multi-point contact mechanics based on the orientation of the posture and the acceptable force margin of the muscles to obtain dynamic contact force; Step S244: Perform adaptive fuzzy inverse kinematics transformation of the robot based on dynamic contact force and orientation to obtain the robot's joint space motion trajectory; Step S245: Based on the robot joint spatial motion trajectory and dynamic contact force, perform data parameter fusion to generate robot auxiliary output parameters.

4. The control method for the nursing robot according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform parameter feature importance learning on the robot's auxiliary output parameters to obtain auxiliary learning parameters; Step S32: Encode instructions based on auxiliary learning parameters to obtain auxiliary parameter instructions; Step S33: Using auxiliary parameter instructions as input, design the control architecture of the nursing robot to build a robot-assisted nursing architecture; Step S34: Send the robot-assisted nursing architecture to the control terminal of the nursing robot to execute the control of the nursing robot.

5. The control method for the nursing robot according to claim 4, characterized in that, Step S33 includes the following steps: Step S331: Take the auxiliary parameter instruction as input, perform instruction element decomposition processing on the auxiliary parameter instruction to obtain a multi-dimensional instruction factor set containing instruction function factor, timing factor and constraint factor; Step S332: Based on the aforementioned multidimensional instruction factor set, deduce the functional interaction coupling relationship of the nursing robot to obtain the functional interaction coupling relationship; Step S333: Learn the policy association of the functional interaction coupling relationship according to the policy gradient algorithm to obtain the interaction coupling association policy; Step S334: Design the control architecture of the nursing robot based on the interaction coupling association strategy to build a robot-assisted nursing architecture.

6. A control system for a nursing robot, characterized in that, For executing the control method of the nursing robot as described in claim 1, the control system of the nursing robot includes: The data acquisition module is used to acquire the patient's posture trajectory when getting up during the nursing process using a 3D camera; and to acquire the muscle tension change signals in the body contact area when the patient gets up during the nursing process using an electromyography (EMG) sensor based on the posture trajectory; and to perform synchronous correlation processing on the muscle tension change signals based on the posture trajectory to obtain posture-related muscle tension change data. The robot output parameter analysis module is used to analyze the muscle group force line imbalance intensity during the posture change process based on the posture-related muscle tension change data. Subsequently, it performs center of gravity offset regression deconstruction to obtain the center of gravity offset regression trend; and generates robot auxiliary output parameters based on the center of gravity offset regression trend. The control architecture design module is used to design the control architecture of the nursing robot by taking the robot-assisted output parameters as input, so as to build the robot-assisted nursing architecture and send it to the control terminal of the nursing robot to execute the control of the nursing robot.

Images