Reconfigurable exoskeleton power-assisted robot, control method, equipment and storage medium

Abstract

The invention discloses a reconfigurable exoskeleton power-assisted robot, a control method, equipment and a storage medium. The control method comprises the following steps that human body motion data are collected in real time through a sensor module; calculating an Euler angle by adopting a quaternion fusion algorithm; predicting a target position based on Kalman filtering; compensating a tracking error through a PID controller; switching a working mode according to the motion state; according to the invention, single interaction limitation is broken through, operation convenience and function expandability are both considered, and efficient man-machine cooperation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of manufacturing industry, and in particular to a reconfigurable exoskeleton power-assisted robot and a control method, device, and storage medium. Background Art

[0002] As labor intensity in industrial manufacturing, logistics and construction continues to increase, industrial workers face the risk of musculoskeletal diseases caused by high-intensity repetitive work for a long time. Although reconfigurable power assist systems, as human functional auxiliary equipment, can effectively alleviate work-related injuries, existing technologies still have significant defects: (1) Functional limitations: They mostly focus on single assistance (such as upper limb lifting or lower limb walking), lack the ability of whole-body coordinated assistance, and cannot adapt to complex scenarios (such as lifting and walking at the same time). (2) Rigid structure and low adaptability: Traditional reconfigurable power assist systems use fixed mechanical structures, which are difficult to adapt to users of different body shapes or dynamic task requirements, and have poor human-machine motion coupling, which can easily cause secondary injuries. (3) Weight and performance contradictions: To ensure structural strength, mainstream products use metal materials (such as aluminum alloy), resulting in a total weight of more than 15kg, which increases the burden on users; lightweight design often sacrifices power output. (4) Intention recognition lag: Motion perception systems that rely on a single sensor have low prediction accuracy and high latency in complex postures, resulting in delayed power assist response. (5) Insufficient interactive experience: Control relies on physical buttons or handheld devices, which cannot free the user's hands, and lacks multimodal interaction channels. Summary of the Invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a reconfigurable exoskeleton-assisted robot and control method, equipment, and storage medium, design a dual-channel efficient human-computer interaction system, and propose a motion intention prediction algorithm based on distributed multi-IMU fusion, which effectively improves the real-time and accuracy of human motion state perception under complex postures, thereby improving the lag of traditional intention recognition; innovating a dual-channel control scheme of voice and APP, breaking through the limitation of single interaction, taking into account both operational convenience and functional expansibility, achieving efficient human-computer collaboration, enhancing the interactive experience, and liberating the user's hands, so as to solve the functional limitations of the existing technology, and realize flexible reconstruction of the exoskeleton function and load. According to the usage scenario, it can be flexibly switched to four working modes: single arm, double arms, lower limbs, and whole body.

[0004] Technical solution: The control method of a reconfigurable exoskeleton-assisted robot described in the present invention comprises the following steps: S1 collects human motion data in real time through sensor modules; S2 uses a quaternion fusion algorithm to calculate Euler angles, including calculating the initial quaternion; updating the iterative quaternion; correcting magnetometer fusion and yaw drift; and converting the updated quaternion into roll, pitch, and yaw angle outputs. S3 predicts target position based on Kalman filtering; includes constructing state equations and observation equations; calculating the predicted state covariance matrix; and updating state estimates and covariance using Kalman gain. S4 compensates for tracking error through PID controller; S5 switches working modes according to the motion status.

[0005] Furthermore, step S1 is specifically as follows: using a distributed nine-axis IMU network to collect roll angle, pitch angle, heading angle data and three-axis acceleration data of multiple parts of the human body in real time; and performing sensor initialization, data reading and preprocessing based on the collected data.

[0006] Furthermore, in step S2, the quaternion fusion algorithm includes the following steps: S21 calculates initial quaternion: calculates initial attitude angle based on accelerometer and magnetometer data and converts it into initial quaternion; the attitude angle includes roll angle, pitch angle, and yaw angle; S22 quaternion update iteration: uses the Mahony complementary filtering algorithm to fuse accelerometer and gyroscope data to suppress drift; includes acceleration normalization, attitude error calculation, and PI controller correction; S23 Correction of Magnetometer Fusion and Yaw Drift: Using magnetometer data to provide an absolute heading reference, the normalized magnetic field vector is converted to the navigation coordinate system, and the reference heading angle is calculated to correct for gyroscope integration drift; S24 outputs Euler angles: Convert the updated quaternion into roll, pitch, and yaw angles for output. The formula is as follows: ; ; ; in, is the roll angle, is the pitch angle, is the yaw angle; is a quaternion.

[0007] Furthermore, in step S3, the Kalman filter prediction includes the following steps: S31 constructs the state equation and observation equation: ; ; in, is the input signal, is the state variable, is the process noise, For the effect The state transition model on is acting on the controller vector The input control model on is the observed variable, is the observation noise, is the observation model; S32 prediction step: Calculate the predicted state covariance matrix, the formula is: ; in, represents the covariance of the forecast state estimate (a measure of forecast uncertainty), is the process noise covariance matrix, S33 update: Use Kalman gain to update state estimation and covariance, the formula is: ; ; ; in, represents the Kalman gain, This is the estimated value of the covariance matrix of the state estimate at time step k based on the previous time step k-1. It represents the estimated value of the covariance matrix of the state estimate at time step k-1 for time step The uncertainty of the state estimate, is the observation matrix, is the covariance matrix of the observation noise, is the state estimate update, is the time step The observed value of is the covariance update, is the identity matrix.

[0008] Furthermore, in step S4, the PID controller compensation includes the following steps: S41 defines the error variable : in Predict the position for the motor, is the actual position of the motor; S42 calculates the compensation amount Δq(t): ; Among them, the proportional term P is , which is used to quickly respond to the current error change rate and form a "prediction-proportional" dual-loop regulation with the Kalman gain; the integral term I is , which is used to accumulate historical errors and eliminate the steady-state deviation caused by mechanical static friction; the differential term D is , used to predict the error change trend and suppress overshoot during rapid arm swing; S43 parameter tuning: Use the Ziegler-Nichols critical proportionality method to tune PID parameters.

[0009] Furthermore, step S43 is as follows: gradually increase the proportional gain Until the system jitters, then based on the critical increase and oscillation period to calculate the final parameters , / , .

[0010] Furthermore, step S5 is specifically as follows: based on the fused Euler angle data and three-axis acceleration values, combined with the preset motor rotation range threshold, the weights of the three angles are adaptively allocated to identify the current upper and lower limb states of the human body, thereby switching the upper limb following mode, upper limb lifting mode, lower limb walking mode and human body carrying mode.

[0011] The reconfigurable exoskeleton assisted robot described in the present invention is implemented using any of the control methods described, and includes: a backboard with a quick-release locking seat; a battery module and a main control module installed on the front of the backboard; a detachable joint module connected to the backboard via a quick-release locking seat; sensor modules distributed on the detachable joint module and the backboard; a voice module installed on the back of the backboard for receiving voice commands; wherein the main control module controls the movement of the joint module in real time through the sensor module.

[0012] An electronic device according to the present invention includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods when executing the program.

[0013] The computer-readable storage medium of the present invention is characterized in that it stores a computer program, and when the program is executed by a processor, the steps of any one of the methods are implemented.

[0014] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: It utilizes a quick-release reconfigurable joint and load distribution structure design concept, enabling flexible reconfiguration of the power assist system load and the ability to flexibly switch between four operating modes depending on the usage scenario. It also proposes a motion intention prediction algorithm based on distributed multi-IMU fusion, effectively improving the real-time and accuracy of human motion state perception in complex postures. It also designs a dual-channel, efficient human-computer interaction system, innovating a dual-channel control solution for voice and app control, breaking through the limitations of single interaction, balancing operational convenience with functional expandability, and achieving efficient human-computer collaboration. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1is a schematic structural diagram of the reconfigurable exoskeleton power-assisted robot of the present invention (reverse side); Figure 2 2. It is a schematic structural diagram of the reconfigurable exoskeleton power-assisted robot of the present invention (front view); Figure 3 This is a structural exploded diagram of the main control module of the present invention; Figure 4 This is an exploded view of the structure of the battery module of the present invention; Figure 5 This is the data acquisition flow chart of the nine-axis IMU of the present invention; Figure 6 It is a flow chart of the collaborative control algorithm of Kalman filtering and PID optimization; Figure 7 It is the motor control flow chart. DETAILED DESCRIPTION

[0016] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0017] like Figure 5 As shown, the embodiment of the present invention also provides a control method for a reconfigurable exoskeleton power-assisted robot, which combines a collaborative control algorithm with Kalman filtering and PID optimization to drive the coordinated operation of multiple joint motors and realize convenient human-machine collaborative assistance through voice interaction. The method includes the following steps: Step 1: Use the nine-axis IMU for data collection; Step 2: Data fusion: used to generate stable and reliable Euler angle data. It includes the following steps: (21) Initial quaternion calculation The present invention first collects sensor data (accelerometer , magnetometer ) is converted into Euler angles and the initial quaternion is calculated based on this. Specifically, the initial attitude angle (roll angle , pitch angle , yaw angle ) The present invention adopts the following formula to calculate: (1); (2); (3); in ; The present invention calculates the initial quaternion from the above Euler angles [] The formula is as follows: (4); (5); (6); (7); In formula (4)-(7) 、 、 、 ,as well as The formula is as follows: (8); (9); (10); (11); (12); (13); (22) Quaternion iterative update: The present invention uses the Mahony complementary filtering algorithm to iteratively update the quaternion and fuses the accelerometer and gyroscope data to suppress drift, specifically including: Acceleration correction: First, the accelerometer and gyroscope are complementary corrected to normalize the acceleration. The formula is as follows: (14); Next, the present invention calculates the cross product of the accelerometer's gravity vector and the gyroscope's predicted gravity vector to generate an attitude error signal. To effectively integrate and suppress drift, the present invention employs a PI controller (proportional gain KP = 1.5, integral gain KI = 0.005). This controller uses the attitude error signal to rapidly respond to the current error, correcting the gyroscope's immediate deviation in real time. The integral term effectively suppresses the accumulated historical error, ultimately outputting the corrected angular velocity, gyro_correct.

[0018] Quaternion update: This invention uses the first-order Runge-Kutta method with less computational effort to update the quaternion, which only requires the use of the corrected gyroscope angular velocity , whose formula is as follows: (15); in is a quaternion, is the gyroscope angular velocity.

[0019] Step 3: Magnetometer fusion and yaw drift correction: To overcome the significant drift problem of the gyroscope in yaw angle (Yaw), the present invention introduces magnetometer data to provide an absolute heading reference. The magnetometer measures the original magnetic field vector in the body coordinate system The magnetometer data processing flow of the present invention is as follows: (16); Since the carrier attitude will affect the magnetometer measurement value, the present invention uses the current attitude quaternion Calculate the rotation matrix , convert the normalized magnetic field vector to the navigation coordinate system, the formula is as follows: (17); The rotation matrix It is the rotation matrix from the body coordinate system to the navigation coordinate system, which can be calculated by quaternion. The formula is as follows: (18); Reference heading calculation and drift correction: The present invention is based on the converted horizontal component and Calculate the reference heading angle: ,in and This is the horizontal component of the magnetic field in the navigation coordinate system. The gyroscope then calculates attitude by integrating the angular velocity, but this can cause drift, particularly in the yaw angle. The absolute heading information provided by the magnetometer can be used to correct for this drift. Finally, the updated quaternion is converted to Euler angles.

[0020] The absolute heading information is used by the present invention to correct the yaw angle drift caused by gyroscope integration, thereby significantly improving the long-term stability of the yaw angle.

[0021] Final attitude output: After completing the fusion update, the present invention converts the quaternion into Euler angle output:

[0022] (19);

[0023] (20);

[0024] (twenty one);

[0025] in is the roll angle, is the pitch angle, is the yaw angle.

[0026] The data fusion algorithm of the present invention achieves high stability and reliability of Euler angle data, providing an accurate attitude reference for high-precision control of the motor.

[0027] Step 4: Collaborative control algorithm based on Kalman filtering and PID optimization: The present invention uses the aforementioned nine-axis gyroscope to collect Euler angle data from five body position sensors, obtaining roll, pitch, and heading angle data as well as acceleration data in the x, y, and z directions for the five body parts. The brushless motors in the upper limbs are controlled by two nine-axis gyroscopes, while the brushless motors in the lower limbs are controlled by three nine-axis gyroscopes on the back and lower limbs.

[0028] Kalman Prediction (State Estimation): The present invention uses the Kalman filter algorithm to predict the target rotational position of the motor. Kalman filtering is an efficient time-domain recursive estimation algorithm that only requires the state estimate value at the previous moment and the current observation value to calculate the current optimal estimate, without storing historical information.

[0029] Algorithm construction: The present invention constructs the state equation and observation equation of the system and sets the initial state. The formula is as follows:

[0030] (twenty two); (twenty three); in is the input signal, is the state variable, is the process noise, For the effect The state transition model on is acting on the controller vector The input control model on is the observed variable, is the observation noise, is the observation model.

[0031] Prediction steps: In order to estimate the error a posteriori, the present invention needs to calculate the covariance error matrix to measure the accuracy of the estimated value. The formula is as follows: (twenty four); in represents the covariance of the forecast state estimate (a measure of forecast uncertainty), is the process noise covariance matrix.

[0032] Update steps: The present invention calculates the current state estimate based on the current observations and predictions. This estimate is more accurate because it takes the current observations into account. The error covariance matrix of the state estimate is calculated using the error covariance matrix calculated in the prediction step, the observation noise covariance matrix, and the Kalman gain. The updated formula is as follows: (25); (26); (27); From the above formula, we can know that the Kalman gain In fact, it represents the ratio of model prediction error (Predicted error) to measurement error (Measurement error) in the process of optimal state estimation (as shown below), that is, .when When the prediction error is 0, the state value of the system depends entirely on the prediction value; and when When , that is, the measurement error is 0, the state value of the system depends entirely on the observation value.

[0033] PID real-time feedback compensation: Building on Kalman prediction, this invention incorporates a PID controller to create a closed-loop feedback loop, achieving precise motor position tracking through real-time error correction. Its control logic complements the "prediction-correction" mechanism of the Kalman filter: while the Kalman filter primarily addresses system model uncertainties (such as sensor noise and modeling errors), the PID controller focuses on suppressing real-time tracking errors caused by external disturbances, load variations, and mechanical static friction.

[0034] The present invention defines the error variable for PID calculation, and the formula is as follows: (28); in Predict the position for the motor, The purpose is to calculate the compensation amount through the PID algorithm described below. , used to suppress the Kalman prediction, the formula is as follows: (29); The proportional term P is , which is used to quickly respond to the current error change rate and form a "prediction-proportional" dual-loop regulation with the Kalman gain; the integral term I is , which is used to accumulate historical errors and eliminate the steady-state deviation caused by mechanical static friction; the differential term D is , which is used to predict the error change trend and suppress overshoot during rapid arm swinging.

[0035] Parameter tuning: This invention uses the Ziegler-Nichols critical proportionality method to perform preliminary tuning of PID parameters. Turn off the function of the proportional term and keep only the function of the proportional term. Start from a small value of 1.0 and gradually increase it to observe the response of the system. Gradually increase until the motor output begins to produce obvious jitter. Then, according to the Ziegler-Nichols empirical formula, Among them used and Calculate PID parameters suitable for the system, , The time is 10ms, and finally it is adjusted according to the actual motor's upper body effect, and then the predicted value of the motor and the compensation value are weighted to generate the final control signal and input it into the motor driver to drive the brushless motor to accurately rotate to the target position. The flow chart of the collaborative control algorithm of Kalman filter and PID optimization is as follows Figure 6 shown.

[0036] The combined Kalman filter and PID control algorithm of this invention overcomes the limitations of single control methods through a clear division of responsibilities (KF predicts global optimality, PID compensates for local errors) and an efficient prediction-compensation closed-loop coordination mechanism. It not only significantly improves the accuracy and speed of motor position tracking, but more importantly, demonstrates excellent stability and robustness in the face of multiple interference sources such as sensor noise, model uncertainty, load variations, and mechanical friction. This provides a reliable, smooth, and responsive execution control core for highly dynamic applications such as exoskeletons.

[0037] Step 5: Based on the fused Euler angle data and triaxial acceleration values, the present invention adaptively assigns weights to the three angles in combination with the preset motor rotation range threshold to identify the current state of the upper and lower limbs of the human body, thereby switching between upper limb following mode, upper limb lifting mode, lower limb walking mode, and human carrying mode. The specific implementation process is as follows: The motor's zero point value is set based on the initial vertical arm position of the human body after the main control is turned on. The weighted value of the upper limb gyroscope is used as the base point for motor rotation. The weighted value of the gyroscope is recorded according to the maximum amplitude of actual human limb rotation and matched with the motor rotation value. Finally, the motion state is determined by combining the values ​​of the three-axis acceleration. The motion modes include upper limb following mode, upper limb lifting mode, lower limb walking mode, lower limb rising mode, and human carrying mode. The details are as follows: First, after the main control is powered on, when the motor is in the upper limb following mode and the lower limb walking mode, the three-axis acceleration value of the gyroscope is recorded every 10ms. When the latest acceleration value is greater than the last recorded value, the upper limb following mode and the lower limb walking mode of the motor are switched to the limb prediction mode, so that the motor drives the human body to move in the predicted direction. The present invention processes the data processed by the five gyroscopes through the main control, and judges the current state of the human body in turn to change the control mode of the motor. It also has a force unloading function. When the current of the motor exceeds the threshold (20A), it will protect the human body and enter the force unloading state. Upper limb following mode and lower limb walking mode: When the exoskeleton robot is powered on, the upper limbs are in upper limb following mode by default, and the lower limbs are in lower limb walking mode by default. In this mode, the distributed nine-axis IMU network The exoskeleton captures human motion intentions in real time. The ESP32 main control unit and STM32 coprocessor run Kalman filtering and PID optimization algorithms to drive four brushless joint motors (peak torque 12N·m). This accurately predicts user intentions and achieves 10ms-level low-latency collaborative assistance, ensuring that the exoskeleton's assistance effect is consistent with the pedestrian's motion needs.

[0038] Upper limb lifting mode: When the user lifts a heavy object to a certain position in upper limb follow mode, the mode can be changed to upper limb lift mode through the voice control module or mobile phone APP. In this mode, the position proportional gain KP and speed differential gain KD will be increased synchronously. A single arm can stably provide a peak power assist effect of 60N, and a double arm can provide 120N to help maintain the current position of the upper limb and reduce the burden on the upper limb. It is suitable for use scenarios where long-term lifting operations are performed, such as photographers or workers. Human transport mode: When the gyroscope on the back detects that the body's leaning angle reaches the set leaning angle threshold, the gyroscope values ​​of the upper and lower limbs are detected at the same time. Whether it reaches the bending state At this time, by judging the three-axis acceleration of the upper and lower limbs and Greater than the acceleration value at the previous moment , When the human body is detected to be back to normal standing state, the lower limb motor enters the human body carrying mode, and the position proportional gain KP and the speed differential gain KD are increased synchronously. The single leg can stably provide a peak power assist effect of 85N, and the single arm can stably provide a peak power assist effect of 60N to help carry heavy objects and reduce the burden on the whole body. It is suitable for use scenarios where people are engaged in heavy object carrying operations for a long time, such as porters. The overall motor control flow chart is as follows Figure 7 shown.

[0039] like Figure 1-Figure 2 As shown, an embodiment of the present invention provides a reconfigurable exoskeleton-assisted robot that can be flexibly switched according to usage scenarios, achieving reconfiguration. The robot comprises a backboard 108, a battery module 104, a main control module 105, a voice module 202, a sensor module (including a first sensor 102, a second sensor 107, a third sensor 203, and a fourth sensor 204), and a detachable joint module (including an upper left joint 101, an upper right joint 106, a lower left joint 103, and a lower right joint 109). The first sensor 102 and the second sensor 107 are connected to the upper left joint 101 and the upper right joint 106, respectively; the third sensor 203 and the fourth sensor 204 are connected to the lower left joint 103 and the lower right joint 109, respectively. The first sensor 102, the third sensor 203, and the fourth sensor 204 have integrated gyroscopes and geomagnetic sensors.

[0040] The backboard 108 is responsible for installing the battery module 104, the main control module 105, the upper left joint 101, the upper right joint 106, the lower left joint 103, and the lower right joint 109. The backboard 108 includes three base plates, two shoulder joint module locking seats, and two leg joint module locking seats, which are responsible for the installation and adjustment of the corresponding joint modules. The connection device of the upper left joint 101, the upper right joint 106, the lower left joint 103, the lower right joint module 109 and the backboard 108 is a quick-release locking seat. When installing the corresponding joint, the joint needs to be inserted into the quick-release locking seat. The upper end of the quick-release locking seat adopts a slotted locking tube, which is matched with a screw-type locking terminal. After the wearer inserts the joint module into the locking seat, he tightens the screw-type locking terminal to quickly complete the installation of the corresponding joint module. like Figure 4 As shown, the battery module 104 includes: a power module fixing module 401, a power module bottom plate 402, a lithium battery 403, and a power module upper cover 404, which is responsible for providing power for the entire exoskeleton robot system; like Figure 3 As shown, the main control module 105 is responsible for controlling the movement of the system, and includes: a main control base plate 301 , a main control circuit board 302 , a main control support plate 303 , a screen base plate 304 , a main control upper cover plate 305 , and a power button 306 .

[0041] The voice module 202 includes a voice board, an external microphone, a voice module upper cover, and a voice module lower cover. The voice module can recognize and analyze the user's voice commands, making it easier to use the exoskeleton without having to use both hands to control the operation.

[0042] The upper left joint 101, the upper right joint 106, the lower left joint 103, and the lower right joint 109 include a joint module cover, a joint module base, a needle bearing, a carbon fiber tube, a carbon fiber tube locking seat, a servo motor, a servo motor output connector, an output connecting carbon fiber tube, a carbon fiber tube sleeve, a binding piece, and a binding piece connector; the carbon fiber tubes in the upper left joint 101, the upper right joint 106, the lower left joint 103, and the lower right joint 109 are installed in the joint module base through needle bearings and fixed through a carbon fiber tube locking seat to provide the shoulder with inward and outward freedom of movement, and the servo motor The machine is fixed to the joint module base by screws, the joint module upper cover is fixed to the joint module bottom cover by screws, the servo motor output connector is connected to the servo motor by six screws, the output connection carbon fiber tube is connected to the servo motor output connector by screws, the binding connector is connected to the output connection carbon fiber tube by screws, the binding is connected to the binding connector by sutures, the first sensor 102, the second sensor 107, the third sensor 203, and the fourth sensor 204 are connected to the binding by sutures, and the length of the binding can be freely adjusted according to one's own body parameters.

Claims

1. A control method for a reconfigurable exoskeleton-assisted robot, characterized in that: The following steps are involved: S1 collects human motion data in real time through sensor modules; S2 uses a quaternion fusion algorithm to calculate Euler angles, including calculating the initial quaternion; updating the iterative quaternion; correcting magnetometer fusion and yaw drift; and converting the updated quaternion into roll, pitch, and yaw angle outputs. S3 predicts the target position based on Kalman filtering; Including constructing the equation of state and observation equation; Calculate the predicted state covariance matrix; Update state estimates and covariances using Kalman gain; S4 compensates for tracking error through PID controller; S5 switches working modes according to the motion status.

2. The control method of a reconfigurable exoskeleton-assisted robot according to claim 1, characterized in that: Step S1 is as follows: using a distributed nine-axis IMU network to collect roll angle, pitch angle, heading angle data and three-axis acceleration data of multiple parts of the human body in real time; and performing sensor initialization, data reading and preprocessing based on the collected data.

3. The control method of a reconfigurable exoskeleton-assisted robot according to claim 1, characterized in that: In step S2, the quaternion fusion algorithm includes the following steps: S21 calculates initial quaternion: calculates initial attitude angle based on accelerometer and magnetometer data and converts it into initial quaternion; the attitude angle includes roll angle, pitch angle, and yaw angle; S22 quaternion update: uses the Mahony complementary filtering algorithm to fuse accelerometer and gyroscope data to suppress drift; includes acceleration normalization, attitude error calculation, and PI controller correction; S23 Correction of Magnetometer Fusion and Yaw Drift: Using magnetometer data to provide an absolute heading reference, the normalized magnetic field vector is converted to the navigation coordinate system, and the reference heading angle is calculated to correct for gyroscope integration drift; S24 converts the updated quaternion into roll, pitch, and yaw angles, which are output as Euler angles. The formula is as follows: ； ； ； in, is the roll angle, is the pitch angle, is the yaw angle; is a quaternion.

4. The control method of a reconfigurable exoskeleton-assisted robot according to claim 1, characterized in that: In step S3, the Kalman filter prediction includes the following steps: S31 constructs the state equation and observation equation: ； ； in, is the input signal, is the state variable, is the process noise, For the effect The state transition model on is acting on the controller vector The input control model on is the observed variable, is the observation noise, is the observation model; S32 prediction step: Calculate the predicted state covariance matrix, the formula is: ； in, represents the covariance of the predicted state estimate, is the process noise covariance matrix, S33 update: Use Kalman gain to update state estimation and covariance, the formula is: ； ； ； in, represents the Kalman gain, This is the estimated value of the covariance matrix of the state estimate at time step k based on the previous time step k-1, which means the estimated value of the covariance matrix of the state estimate at time step k-1 for time step The uncertainty of the state estimate, is the observation matrix, is the covariance matrix of the observation noise, is the state estimate update, is the time step The observed value of is the covariance update, is the identity matrix.

5. The control method of a reconfigurable exoskeleton-assisted robot according to claim 1, characterized in that: In step S4, PID controller compensation includes the following steps: S41 defines the error variable : in Predict the position for the motor, is the actual position of the motor; S42 calculates the compensation amount Δq(t): ； Among them, the proportional term P is , which is used to quickly respond to the current error change rate and form a "prediction-proportional" dual-loop regulation with the Kalman gain; the integral term I is , which is used to accumulate historical errors and eliminate the steady-state deviation caused by mechanical static friction; the differential term D is , used to predict the error change trend and suppress overshoot during rapid arm swing; S43 parameter tuning: Use the Ziegler-Nichols critical proportionality method to tune PID parameters.

6. The control method of a reconfigurable exoskeleton-assisted robot according to claim 5, characterized in that: Step S43 is as follows: gradually increase the proportional gain Until the system jitters, then based on the critical increase and oscillation period to calculate the final parameters , / , .

7. The control method of a reconfigurable exoskeleton-assisted robot according to claim 1, characterized in that: Step S5 is specifically as follows: Based on the fused Euler angle data and three-axis acceleration values, combined with the preset motor rotation range threshold, the weights of the three angles are adaptively allocated to identify the current upper and lower limb states of the human body, thereby switching the upper limb following mode, upper limb lifting mode, lower limb walking mode and human body carrying mode.

8. A reconfigurable exoskeleton-assisted robot, characterized in that: The control method according to any one of claims 1 to 7 is used for implementation, comprising: a backpack backboard (108) provided with a quick-release locking seat; a battery module (104) and a main control module (105) installed on the front of the backpack backboard (108); a detachable joint module connected to the backpack backboard via a quick-release locking seat; sensor modules distributed on the detachable joint module and the backpack backboard (108); a voice module (202) installed on the back of the backpack backboard (108) for receiving voice commands; wherein the main control module (105) controls the movement of the joint module in real time through the sensor module.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that A computer program is stored, and when the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images