Exoskeleton robot cooperative control method and device and storage medium

Abstract

The invention provides an exoskeleton robot cooperative control method and device and a storage medium, and belongs to the technical field of robot control. The method comprises the following steps: acquiring an original surface electromyogram signal, an original joint angle signal, an original joint angular velocity signal and an original human body motion acceleration signal through a sensor group arranged at a preset part of a human body; the method effectively captures the associated information and dynamic change characteristics among the multi-source signals, significantly improves the accuracy of human motion intention recognition, reduces the signal processing time, enhances the adaptability of cooperative control, meets the use requirements of different users in different scenes, and improves the user experience. The stability and the accuracy of movement of the exoskeleton robot are ensured, the risk of injury to a human body due to movement deviation is reduced, and the use safety is improved.

Description

TECHNICAL FIELD

[0001] The present application mainly relates to the field of robot control technology, and in particular to a kind of exoskeleton robot cooperative control method, device and storage medium. BACKGROUND

[0002] As an equipment capable of assisting and enhancing human movement ability, exoskeleton robot has been widely concerned. One of the core technologies of exoskeleton robot is the cooperative control of its movement with human body. Only when the two are well coordinated, can the exoskeleton robot ensure that it does not interfere with the human body when assisting the movement of the human body, while effectively reducing the burden on the human body and improving the movement efficiency.

[0003] In the prior art, the cooperative control of exoskeleton robot and human body movement mainly has the following problems: Low accuracy of movement intention recognition: At present, most exoskeleton robots judge human movement intention by collecting single or a few biological signals such as surface electromyogram signals and joint angle signals of human body. However, these signals are easily affected by external interference (such as skin contact condition, environmental noise, etc.) and physiological state changes (such as muscle fatigue) of human body, resulting in low accuracy of human movement intention recognition, which further affects the cooperative control effect.

[0004] Response lag of cooperative control: In the existing control method, there is a certain time delay between signal collection, processing and control instruction generation and execution. When the movement state of human body changes rapidly, the exoskeleton robot cannot make corresponding adjustment in time, which is easy to produce movement lag phenomenon, not only affecting the assistance effect, but also possibly causing discomfort or even harm to human body.

[0005] Poor adaptability: The body structure, movement habit and movement ability of different individuals are quite different, and the movement characteristics of the same individual will also change in different movement stages (such as movement starting stage, movement stable stage and movement ending stage). However, the control parameters of existing exoskeleton robots are mostly fixed, which is difficult to adaptively adjust according to individual differences and movement stage changes, resulting in poor adaptability of cooperative control, which cannot meet the use requirements of different users in different scenes. SUMMARY

[0006] The technical problem to be solved by the present application is to provide an exoskeleton robot cooperative control method, device and storage medium to solve the problems of the prior art.

[0007] The technical solution of the present application to solve the above technical problem is as follows: an exoskeleton robot cooperative control method, comprising the following steps: a plurality of original surface electromyography signals obtained by a sensor group arranged at a preset part of a human body, an original joint angle signal corresponding to each of the original surface electromyography signals, an original joint angular velocity signal corresponding to each of the original surface electromyography signals, and an original human motion acceleration signal corresponding to each of the original surface electromyography signals; each of the original surface electromyography signals, the original joint angle signal corresponding to each of the original surface electromyography signals, the original joint angular velocity signal corresponding to each of the original surface electromyography signals, and the original human motion acceleration signal corresponding to each of the original surface electromyography signals is preprocessed to obtain a preprocessed surface electromyography signal corresponding to each of the original surface electromyography signals, a preprocessed joint angle signal corresponding to each of the original surface electromyography signals, a preprocessed joint angular velocity signal corresponding to each of the original surface electromyography signals, and a preprocessed human motion acceleration signal corresponding to each of the original surface electromyography signals; each of the preprocessed surface electromyography signals, the preprocessed joint angle signal corresponding to each of the original surface electromyography signals, the preprocessed joint angular velocity signal corresponding to each of the original surface electromyography signals, and the preprocessed human motion acceleration signal corresponding to each of the original surface electromyography signals is extracted to obtain an original time domain feature corresponding to each of the original surface electromyography signals, an original joint angle feature corresponding to each of the original surface electromyography signals, an original joint angular velocity feature corresponding to each of the original surface electromyography signals, and an original acceleration feature corresponding to each of the original surface electromyography signals; a classification model is constructed, and all of the original time domain features, all of the original joint angle features, all of the original joint angular velocity features, and all of the original acceleration features are fused and analyzed by the classification model to obtain human motion intention category information; a plurality of interaction force signals are obtained by a force sensor arranged on an exoskeleton robot; all of the original time domain features, all of the original joint angle features, all of the original joint angular velocity features, all of the original acceleration features, and all of the interaction force signals are subjected to parameter analysis to obtain target exoskeleton robot control parameters; a motion control instruction is generated according to the human motion intention category information and the target exoskeleton robot control parameters, and the exoskeleton robot is controlled to move according to the motion control instruction.

[0008] Another technical solution of the present application to solve the above technical problems is as follows: an exoskeleton robot cooperative control device, comprising: The first signal obtaining module is configured to obtain a plurality of original surface electromyography signals, original joint angle signals corresponding to the original surface electromyography signals, original joint angular velocity signals corresponding to the original surface electromyography signals, and original human motion acceleration signals corresponding to the original surface electromyography signals through a sensor group arranged at a preset part of a human body; The preprocessing module is configured to preprocess the original surface electromyography signals, the original joint angle signals corresponding to the original surface electromyography signals, the original joint angular velocity signals corresponding to the original surface electromyography signals, and the original human motion acceleration signals corresponding to the original surface electromyography signals respectively to obtain preprocessed surface electromyography signals corresponding to the original surface electromyography signals, preprocessed joint angle signals corresponding to the original surface electromyography signals, preprocessed joint angular velocity signals corresponding to the original surface electromyography signals, and preprocessed human motion acceleration signals corresponding to the original surface electromyography signals. The feature extraction module is configured to extract original time domain features corresponding to the original surface electromyography signals, original joint angle features corresponding to the original surface electromyography signals, original joint angular velocity features corresponding to the original surface electromyography signals, and original acceleration features corresponding to the original surface electromyography signals from the preprocessed surface electromyography signals, the preprocessed joint angle signals corresponding to the original surface electromyography signals, the preprocessed joint angular velocity signals corresponding to the original surface electromyography signals, and the preprocessed human motion acceleration signals corresponding to the original surface electromyography signals respectively. The fusion analysis module is configured to construct a classification model, and perform fusion analysis on all the original time domain features, all the original joint angle features, all the original joint angular velocity features, and all the original acceleration features through the classification model to obtain human motion intention category information. The second signal obtaining module is configured to obtain a plurality of interaction force signals through force sensors arranged on an exoskeleton robot. The parameter analysis module is configured to perform parameter analysis on all the original time domain features, all the original joint angle features, all the original joint angular velocity features, all the original acceleration features, and all the interaction force signals to obtain target exoskeleton robot control parameters. The control module is configured to generate a motion control instruction according to the human motion intention category information and the target exoskeleton robot control parameters, and control the exoskeleton robot to perform motion according to the motion control instruction.

[0009] Based on the above-mentioned exoskeleton robot cooperative control method, the application further provides an exoskeleton robot cooperative control device.

[0010] Another technical solution solving the above technical problems of the present application is as follows: an exoskeleton robot cooperative control device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, when the processor executes the computer program, the exoskeleton robot cooperative control method as described above is realized.

[0011] Based on the above exoskeleton robot cooperative control method, the present application further provides a computer readable storage medium.

[0012] Another technical solution solving the above technical problems of the present application is as follows: a computer readable storage medium, the computer readable storage medium stores a computer program, when the computer program is executed by a processor, the exoskeleton robot cooperative control method as described above is realized.

[0013] The beneficial effects of the present application are: through preprocessing of the original surface electromyography signal, the original joint angle signal, the original joint angular velocity signal and the original human body motion acceleration signal to obtain the preprocessed surface electromyography signal, the preprocessed joint angle signal, the preprocessed joint angular velocity signal and the preprocessed human body motion acceleration signal, the original time domain feature, the original joint angle feature, the original joint angular velocity feature and the original acceleration feature are extracted from the preprocessed surface electromyography signal, the preprocessed joint angle signal, the preprocessed joint angular velocity signal and the preprocessed human body motion acceleration signal, the human motion intention category information is obtained through fusion analysis of the original time domain feature, the original joint angle feature, the original joint angular velocity feature and the original acceleration feature by the classification model, the target exoskeleton robot control parameter is obtained through parameter analysis of the original time domain feature, the original joint angle feature, the original joint angular velocity feature, the original acceleration feature and the interaction force signal, and the exoskeleton robot is controlled according to the human motion intention category information and the target exoskeleton robot, which effectively captures the correlation information and dynamic change characteristics between multiple source signals, significantly improves the accuracy of human motion intention recognition, reduces the signal processing time, enhances the adaptability of cooperative control, meets the use requirements of different users in different scenarios, ensures the stability and accuracy of exoskeleton robot motion, reduces the risk of injury to the human body due to motion deviation, and improves the use safety. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A flowchart of the exoskeleton robot cooperative control method provided by the embodiment of the present application is shown in the figure; Figure 2 A module block diagram of the exoskeleton robot cooperative control device provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0015] The principles and features of the present application are described below in conjunction with the accompanying drawings, which are provided only for explanation of the present application and are not intended to limit the scope of the present application.

[0016] Figure 1 A flowchart of a kind of exoskeleton robot cooperative control method provided for the embodiment of the present application.

[0017] As shown in Figure 1 A kind of exoskeleton robot cooperative control method, comprising the following steps: S1: obtaining multiple original surface electromyogram signals, original joint angle signals corresponding to each of the original surface electromyogram signals, original joint angular velocity signals corresponding to each of the original surface electromyogram signals and original human motion acceleration signals corresponding to each of the original surface electromyogram signals through the sensor group arranged at the preset part of human body; S2: respectively pre-processing each of the original surface electromyogram signals, original joint angle signals corresponding to each of the original surface electromyogram signals, original joint angular velocity signals corresponding to each of the original surface electromyogram signals and original human motion acceleration signals corresponding to each of the original surface electromyogram signals, to obtain pre-processed surface electromyogram signals corresponding to each of the original surface electromyogram signals, pre-processed joint angle signals corresponding to each of the original surface electromyogram signals, pre-processed joint angular velocity signals corresponding to each of the original surface electromyogram signals and pre-processed human motion acceleration signals corresponding to each of the original surface electromyogram signals; S3: respectively extracting original time domain features corresponding to each of the original surface electromyogram signals, original joint angle features corresponding to each of the original surface electromyogram signals, original joint angular velocity features corresponding to each of the original surface electromyogram signals and original acceleration features corresponding to each of the original surface electromyogram signals from each of the pre-processed surface electromyogram signals, pre-processed joint angle signals corresponding to each of the original surface electromyogram signals, pre-processed joint angular velocity signals corresponding to each of the original surface electromyogram signals and pre-processed human motion acceleration signals corresponding to each of the original surface electromyogram signals; S4: constructing a classification model, and performing fusion analysis on all the original time domain features, all the original joint angle features, all the original joint angular velocity features and all the original acceleration features through the classification model to obtain human motion intention category information; S5: obtaining multiple interaction force signals through the force sensor arranged on the exoskeleton robot; S6: performing parameter analysis on all the original time domain features, all the original joint angle features, all the original joint angular velocity features, all the original acceleration features, and all the interaction force signals to obtain target exoskeleton robot control parameters; S7: generating a motion control instruction according to the human motion intention category information and the target exoskeleton robot control parameters, and controlling the exoskeleton robot to move according to the motion control instruction.

[0018] It should be understood that the surface electromyography signal (sEMG) (i.e., the original surface electromyography signal), the joint angle signal (i.e., the original joint angle signal), the joint angular velocity signal (i.e., the original joint angular velocity signal), and the human motion acceleration signal (i.e., the original human motion acceleration signal) of the human body are collected simultaneously by the sensor group arranged at the key parts of the human body (such as the muscle attachment and joint of the arm and leg) (i.e., the preset parts of the human body). Among them, the surface electromyography sensor is used to collect the electrical signal generated when the muscle contracts, reflecting the activity state of the muscle; the joint angle sensor adopts a high-precision potentiometer or an optical encoder, which is used to measure the rotation angle of the joint; the joint angular velocity sensor acquires the speed of joint rotation by differentiating the joint angle signal or directly using an angular velocity sensor; the acceleration sensor adopts a three-axis acceleration sensor, which is used to collect the acceleration information in the process of human motion, reflecting the dynamic characteristics of human motion.

[0019] Specifically, the feature parameters are extracted from the preprocessed multi-source biological signals (i.e., the preprocessed surface electromyography signal, the preprocessed joint angle signal, the preprocessed joint angular velocity signal, and the preprocessed human motion acceleration signal). For the surface electromyography signal, the integral electromyography value (IEMG), the root mean square value (RMS), the mean absolute value (MAV), and other time domain features (i.e., the original time domain features) are extracted; for the joint angle signal, the maximum angle, the minimum angle, the angle change range, and other features (i.e., the original joint angle features) are extracted; for the joint angular velocity signal, the maximum angular velocity, the minimum angular velocity, the angular velocity change rate, and other features (i.e., the original joint angular velocity features) are extracted; for the acceleration signal, the maximum acceleration, the minimum acceleration, the acceleration change amplitude, and other features (i.e., the original acceleration features) are extracted.

[0020] It should be understood that, according to the identified human motion intention (i.e., human motion intention category information) and the adaptively adjusted cooperative control parameters (i.e., target exoskeleton robot control parameters), the motion control instructions (such as joint rotation angle instructions, joint rotation speed instructions, joint driving torque instructions, etc.) of each joint of the exoskeleton robot are generated; then, the control instructions (i.e., motion control instructions) are sent to the driving module (such as a servo motor driver) of the exoskeleton robot, and each joint of the exoskeleton robot is driven to move according to the control instructions, realizing the cooperation with the human limb movement.

[0021] In the above embodiment, the original surface electromyogram signals, the original joint angle signals, the original joint angular velocity signals, and the original human motion acceleration signals are preprocessed to obtain preprocessed surface electromyogram signals, preprocessed joint angle signals, preprocessed joint angular velocity signals, and preprocessed human motion acceleration signals; the original time domain features, the original joint angle features, the original joint angular velocity features, and the original acceleration features are extracted from the preprocessed surface electromyogram signals, the preprocessed joint angle signals, the preprocessed joint angular velocity signals, and the preprocessed human motion acceleration signals; the human motion intention category information is obtained by fusion analysis of the original time domain features, the original joint angle features, the original joint angular velocity features, and the original acceleration features through the classification model; the target exoskeleton robot control parameters are obtained by parameter analysis of the original time domain features, the original joint angle features, the original joint angular velocity features, the original acceleration features, and the interaction force signals; and the exoskeleton robot is moved according to the human motion intention category information and the target exoskeleton robot control parameters. The correlation information and dynamic change characteristics between multiple sources of signals are effectively captured, the accuracy of human motion intention recognition is significantly improved, the signal processing time is reduced, the adaptability of cooperative control is enhanced, the use requirements of different users in different scenarios are met, the stability and accuracy of exoskeleton robot movement are ensured, the risk of injury to the human body caused by movement deviation is reduced, and the use safety is improved.

[0022] Optionally, as an embodiment of the present application, the process of respectively preprocessing each of the original surface electromyogram signals, the original joint angle signals corresponding to each of the original surface electromyogram signals, the original joint angular velocity signals corresponding to each of the original surface electromyogram signals, and the original human motion acceleration signals corresponding to each of the original surface electromyogram signals to obtain the preprocessed surface electromyogram signals corresponding to each of the original surface electromyogram signals, the preprocessed joint angle signals corresponding to each of the original surface electromyogram signals, the preprocessed joint angular velocity signals corresponding to each of the original surface electromyogram signals, and the preprocessed human motion acceleration signals corresponding to each of the original surface electromyogram signals includes: The original surface electromyogram signals are filtered respectively to obtain filtered surface electromyogram signals corresponding to the original surface electromyogram signals respectively. The filtered surface electromyogram signals are rectified respectively to obtain rectified surface electromyogram signals corresponding to the original surface electromyogram signals respectively. The rectified surface electromyogram signals are integrated respectively by using a sliding window integration algorithm to obtain preprocessed surface electromyogram signals corresponding to the original surface electromyogram signals respectively. The original joint angle signals and original joint angular velocity signals corresponding to the original surface electromyogram signals are smoothed respectively to obtain smoothed joint angle signals corresponding to the original surface electromyogram signals and smoothed joint angular velocity signals corresponding to the original surface electromyogram signals respectively. The smoothed joint angle signals and smoothed joint angular velocity signals corresponding to the original surface electromyogram signals are filtered respectively to obtain preprocessed joint angle signals corresponding to the original surface electromyogram signals and preprocessed joint angular velocity signals corresponding to the original surface electromyogram signals respectively. The original human motion acceleration signals are calibrated respectively to obtain calibrated human motion acceleration signals corresponding to the original surface electromyogram signals respectively. The calibrated human motion acceleration signals are filtered respectively by using a low-pass filter to obtain preprocessed human motion acceleration signals corresponding to the original surface electromyogram signals.

[0023] It should be understood that the collected surface electromyogram signals (i.e. original surface electromyogram signals) are filtered, a 50Hz notch filter is used to remove power frequency interference, and a 10-500Hz band-pass filter is used to remove low-frequency noise and high-frequency noise; then the filtered signals (i.e. filtered surface electromyogram signals) are rectified to convert negative half-cycle signals into positive half-cycle signals; finally, the rectified signals (i.e. rectified surface electromyogram signals) are integrated by using a sliding window integration method to obtain the integral value of the surface electromyogram signals (i.e. preprocessed surface electromyogram signals), which is used for subsequent feature extraction.

[0024] Specifically, the collected joint angle signals (i.e. original joint angle signals) and joint angular velocity signals (i.e. original joint angular velocity signals) are smoothed, a moving average filtering method is used to remove random noise in the signals (i.e. smoothed joint angle signals and smoothed joint angular velocity signals), and the smoothness and accuracy of the signals are ensured.

[0025] Specifically, the acceleration signal (i.e., the original human motion acceleration signal) is subjected to zero drift calibration to remove errors caused by the sensor's own zero drift; then a low-pass filter is used to filter the calibrated acceleration signal (i.e., the calibrated human motion acceleration signal) to remove high-frequency noise and retain useful motion acceleration information.

[0026] In the above embodiments, the original surface electromyography signal, the original joint angle signal, the original joint angular velocity signal, and the original human motion acceleration signal are preprocessed to obtain a preprocessed surface electromyography signal, a preprocessed joint angle signal, a preprocessed joint angular velocity signal, and a preprocessed human motion acceleration signal, effectively capturing the correlation information and dynamic change characteristics between multiple source signals, significantly improving the accuracy of human motion intention recognition, reducing signal processing time, and enhancing the adaptability of collaborative control.

[0027] Optionally, as an embodiment of the present application, the classification model comprises a convolutional neural network, a recurrent neural network, an attention mechanism layer, and a fully connected layer. The process of fusion analysis of all the original time domain features, all the original joint angle features, all the original joint angular velocity features, and all the original acceleration features by the classification model to obtain human motion intention category information comprises: The convolutional neural network is used to perform local feature extraction on each of the original time domain features, the original joint angle features corresponding to each of the original surface electromyography signals, the original joint angular velocity features corresponding to each of the original surface electromyography signals, and the original acceleration features corresponding to each of the original surface electromyography signals, to obtain local time domain features corresponding to each of the original surface electromyography signals, local joint angle features corresponding to each of the original surface electromyography signals, local joint angular velocity features corresponding to each of the original surface electromyography signals, and local acceleration features corresponding to each of the original surface electromyography signals. The recurrent neural network is used to perform dynamic feature extraction on each of the local time domain features, the local joint angle features corresponding to each of the original surface electromyography signals, the local joint angular velocity features corresponding to each of the original surface electromyography signals, and the local acceleration features corresponding to each of the original surface electromyography signals, to obtain dynamic time domain features corresponding to each of the original surface electromyography signals, dynamic joint angle features corresponding to each of the original surface electromyography signals, dynamic joint angular velocity features corresponding to each of the original surface electromyography signals, and dynamic acceleration features corresponding to each of the original surface electromyography signals. The target fusion feature is obtained by fusing all the dynamic time domain features, all the dynamic joint angle features, all the dynamic joint angular velocity features and all the dynamic acceleration features through the attention mechanism layer. The human motion intention category information is obtained by classifying the target fusion feature through the full connection layer.

[0028] It should be understood that the extracted multi-source features (i.e., the original time domain features, the original joint angle features, the original joint angular velocity features and the original acceleration features) are fused and classified by using the deep learning model based on the attention mechanism (i.e., the classification model).

[0029] Specifically, first, the features (i.e., the original time domain features, the original joint angle features, the original joint angular velocity features and the original acceleration features) of each single-source signal are locally extracted by the CNN model (i.e., the convolutional neural network), so as to capture the local key information in the signal; then, the local features (i.e., the local time domain features, the local joint angle features, the local joint angular velocity features and the local acceleration features) output by the CNN model are input into the RNN model (i.e., the recurrent neural network), so as to capture the dynamic correlation information of the multi-source signal features changing with time by using the time sequence modeling capability of the RNN model (i.e., the recurrent neural network); at the same time, the attention mechanism is introduced, so as to give different weights to the features of different source signals and different time steps, and highlight the feature information which contributes more to the human motion intention recognition; finally, the fused features (i.e., the target fusion feature) are classified by the full connection layer, so as to output the motion intention category of the human (such as walking, running, going up and down the stairs, lifting the hand, bending the waist, etc.) (i.e., the human motion intention category information).

[0030] In the above embodiment, the human motion intention category information is obtained by fusing and analyzing the original time domain features, the original joint angle features, the original joint angular velocity features and the original acceleration features through the classification model, the adaptability of the collaborative control is enhanced, the use requirements of different users in different scenarios are met, the stability and accuracy of the motion of the exoskeleton robot are ensured, the risk of injury to the human body caused by motion deviation is reduced, and the use safety is improved.

[0031] Optionally, as one embodiment of the present application, the process of performing parameter analysis on all the original time domain features, all the original joint angle features, all the original joint angular velocity features, all the original acceleration features and all the interaction force signals to obtain the target exoskeleton robot control parameter includes: The original time domain features are respectively calculated by the difference value of the standard time domain features in the pre-constructed individual motion feature library, so as to obtain the time domain feature deviation value corresponding to each original surface electromyogram signal. respectively, to obtain joint angle feature deviation values corresponding to each of the original surface electromyography signals; respectively, to obtain joint angle velocity feature deviation values corresponding to each of the original surface electromyography signals; respectively, to obtain acceleration feature deviation values corresponding to each of the original surface electromyography signals; a plurality of standard acceleration values are extracted from the pre-constructed individual motion feature library; stability evaluation results are obtained by performing stability evaluation analysis on all the original acceleration features, all the interactive force signals, and all the standard acceleration values; initial exoskeleton robot control parameters are imported, and the initial exoskeleton robot control parameters are adjusted according to a fuzzy PID control algorithm, the stability evaluation results, all the time domain feature deviation values, all the joint angle feature deviation values, all the joint angle velocity feature deviation values, and all the acceleration feature deviation values, to obtain target exoskeleton robot control parameters.

[0032] It should be understood that when a user first uses an exoskeleton robot, a user's individual motion feature model (i.e., a pre-constructed individual motion feature library) is established by collecting the user's multi-source biological signals under different standard motion modes (such as standard walking, standard running, etc.) and combining the motion parameters of the exoskeleton robot (such as joint driving torque, motion speed, etc.). The model includes biological signal feature reference values, motion parameter reference values, and the correlation between biological signals and motion parameters of the user under different motion modes.

[0033] Specifically, in the process of the user using the exoskeleton robot for motion, the current collected and pre-processed multi-source biological signal features (i.e., original time domain features, original joint angle features, original joint angle velocity features, and original acceleration features) are compared with the reference features (i.e., standard time domain features, standard joint angle features, standard joint angle velocity features, and standard acceleration features) in the individual motion feature model in real time, and feature deviation values (i.e., time domain feature deviation values, joint angle feature deviation values, joint angle velocity feature deviation values, and acceleration feature deviation values) are calculated. At the same time, the interactive force signals between the exoskeleton and the human body are collected by the force sensor of the exoskeleton robot, and the current motion stability and fatigue degree of the human body are evaluated in combination with the acceleration signal (i.e., the original acceleration feature).

[0034] It should be understood that according to the real-time motion state evaluation result (i.e. the stability evaluation result), the fuzzy PID control algorithm is used to adaptively adjust the cooperative control parameters (such as the driving torque coefficient of the exoskeleton joint, the motion speed following coefficient, etc.) (i.e. the initial exoskeleton robot control parameters). When the characteristic deviation value (i.e. the time domain characteristic deviation value, the joint angle characteristic deviation value, the joint angular velocity characteristic deviation value, and the acceleration characteristic deviation value) is small, it indicates that the human motion state is close to the reference state, and the adjustment amplitude of the control parameters (i.e. the initial exoskeleton robot control parameters) is appropriately reduced to maintain the stability of the exoskeleton motion; when the characteristic deviation value (i.e. the time domain characteristic deviation value, the joint angle characteristic deviation value, the joint angular velocity characteristic deviation value, and the acceleration characteristic deviation value) is large or the human body appears fatigue signs, the adjustment amplitude of the control parameters (i.e. the initial exoskeleton robot control parameters) is increased to improve the assisting strength of the exoskeleton to the human motion and reduce the burden of the human body.

[0035] In the above embodiment, the target exoskeleton robot control parameters are obtained by performing parameter analysis on the original time domain characteristics, the original joint angle characteristics, the original joint angular velocity characteristics, the original acceleration characteristics, and the interaction force signals, which can evaluate the current motion stability and fatigue degree of the human body, improve the assisting strength of the exoskeleton to the human motion, and reduce the burden of the human body.

[0036] Optionally, as an embodiment of the present application, the process of performing stability evaluation analysis on all the original acceleration characteristics, all the interaction force signals, and all the standard acceleration values to obtain a stability evaluation result comprises: extracting an interaction force amplitude corresponding to each of the original surface electromyography signals from each of the interaction force signals; calculating the standard deviation of all the interaction force amplitudes to obtain an interaction force amplitude standard deviation; calculating the average value of all the interaction force amplitudes to obtain an interaction force amplitude average value; calculating the interaction force amplitude standard deviation and the interaction force amplitude average value by the first formula to obtain an interaction force fluctuation coefficient, the first formula being: , wherein, the interaction force fluctuation coefficient is, the interaction force amplitude standard deviation is, the interaction force amplitude average value is; extracting an original acceleration value corresponding to each of the original surface electromyography signals from each of the original acceleration characteristics; calculating the standard deviation of all the original acceleration values to obtain an original acceleration standard deviation; The acceleration trajectory deviation value is calculated by a second formula on all the original acceleration values and all the standard acceleration values, the second formula being: , wherein, is the acceleration trajectory deviation value, is the number of original acceleration values, is the original acceleration value, is the standard acceleration value; is the standard acceleration value; is the standard acceleration value; The stability score is calculated by a third formula on the interaction force fluctuation coefficient, the original acceleration standard deviation and the acceleration trajectory deviation value, the third formula being: , wherein, is the stability score, is the interaction force fluctuation coefficient, is the original acceleration standard deviation, is the acceleration trajectory deviation value; If the stability score is greater than or equal to a preset first threshold value, a first preset evaluation result is taken as the stability evaluation result; If the stability score is less than the preset first threshold value and greater than or equal to a preset second threshold value, a second preset evaluation result is taken as the stability evaluation result; If the stability score is less than the preset second threshold value, a third preset evaluation result is taken as the stability evaluation result.

[0037] Preferably, the preset first threshold value can be 0.8 and the preset second threshold value can be 0.5.

[0038] It should be understood that the FFC (i.e. the interaction force fluctuation coefficient) reflects the fluctuation degree of the interaction force — when the normal stable movement (such as walking at a constant speed), the human body muscle control is stable, the fluctuation of the exoskeleton auxiliary force is small, and the FFC (i.e. the interaction force fluctuation coefficient) is usually less than 0.15; when the stability of the human body decreases (such as muscle control disorder caused by fatigue), the interaction force will appear irregular fluctuation, and the FFC (i.e. the interaction force fluctuation coefficient) will significantly increase (such as more than 0.3).

[0039] Specifically, when the movement is stable, the human body side sway is small, and the lateral ASD (i.e. the original acceleration standard deviation) of the waist is usually less than 0.5 m / s²; when the stability of the human body decreases (such as standing instability and gait disorder), the lateral ASD (i.e. the original acceleration standard deviation) will significantly increase (such as more than 1.2 m / s²), and the balance auxiliary mechanism of the exoskeleton needs to be triggered.

[0040] Specifically, when the motion is stable, the real-time acceleration trajectory is highly consistent with the standard trajectory, and the ATD (i.e., acceleration trajectory deviation value) is usually less than 0.2 m / s²; when the motion state of the human body is abnormal (such as sudden deceleration or loss of control in turning), the ATD (i.e., acceleration trajectory deviation value) will exceed 0.5 m / s², indicating a stability risk.

[0041] It should be understood that SS (i.e., stability score) ≥ 0.8 is "stable", 0.5 ≤ SS (i.e., stability score) < 0.8 is "basically stable", and SS (i.e., stability score) < 0.5 is "unstable" (active intervention of the exoskeleton is required for balance control).

[0042] In the above embodiment, the stability evaluation result is obtained by performing stability evaluation analysis on the original acceleration characteristics, the interaction force signal, and the standard acceleration value, which can evaluate the current motion stability and fatigue degree of the human body, improve the assistance of the exoskeleton to the human motion, and reduce the burden on the human body.

[0043] Optionally, as another embodiment of the present application, the present application relates to the technical field of exoskeleton robots. The present application can be widely applied to medical rehabilitation exoskeleton robots, industrial auxiliary exoskeleton robots, military load-bearing exoskeleton robots, and other scenes, realizing precise and efficient cooperation between the exoskeleton robot and the human limb motion, and improving the use performance and user experience of the exoskeleton robot.

[0044] Optionally, as another embodiment of the present application, the present application aims to solve the technical problems of low motion intention recognition accuracy, lagging response of cooperative control, and poor adaptability in the existing exoskeleton robot and human limb motion cooperative control method.

[0045] Optionally, as another embodiment of the present application, the present application further comprises: The motion feedback signals (such as actual joint angle, actual joint speed, actual driving torque, etc.) of each joint of the exoskeleton robot are received in real time, and compared with the control command to calculate the motion error; if the motion error exceeds the preset threshold, the control command is corrected in real time to ensure the cooperative accuracy of the exoskeleton robot and the human limb motion.

[0046] Optionally, as another embodiment of the present application, compared with the prior art, the present application has the following beneficial effects: a. Improve the accuracy of motion intention recognition: The present application can reflect the motion state of the human body from multiple dimensions by collecting multi-source biological signals (surface electromyogram, joint angle signal, joint angular velocity signal, acceleration signal), avoiding the problem of single signal being easily disturbed; at the same time, the deep learning model based on attention mechanism is used to fuse and classify the multi-source features, which can effectively capture the correlation information and dynamic change characteristics between multi-source signals, significantly improving the accuracy of human motion intention recognition.

[0047] b. Reduce the response lag of cooperative control: In the signal preprocessing process, efficient filtering and feature extraction algorithms are used to reduce the signal processing time; in the control command generation and execution link, through real-time reception of motion feedback signals and error correction, the rapid adjustment of control commands is realized, effectively reducing the response lag of cooperative control, ensuring that the exoskeleton robot can follow the changes of human motion in time.

[0048] c. Enhance the adaptability of cooperative control: The present application can fully adapt to the individual differences of different users and the changes of motion characteristics of the same user in different motion stages by establishing a user individual motion characteristic model and using fuzzy PID control algorithm to adaptively adjust the cooperative control parameters according to the real-time motion state evaluation results, significantly enhancing the adaptability of cooperative control and meeting the use requirements of different users in different scenarios.

[0049] d. Improve user experience and safety: Since the present application can realize precise and efficient cooperation between exoskeleton robot and human limb motion, it avoids the phenomenon of motion interference, and can adjust the assisting force according to the degree of human fatigue, effectively reducing the burden on the human body and improving the comfort of user use; in addition, through real-time motion error correction, the stability and accuracy of the exoskeleton robot motion are ensured, the risk of injury to the human body caused by motion deviation is reduced, and the use safety is improved.

[0050] Optionally, as another embodiment of the present application, the present application specifically includes: I. Experimental equipment and parameter setting (1) Exoskeleton robot: Select lower limb exoskeleton robot, which contains three key joints of hip joint, knee joint and ankle joint, each joint is driven by servo motor, equipped with high-precision joint angle sensor (resolution is 0.01°), joint angular velocity sensor (measurement range is -300° / s~300° / s) and force sensor (measurement range is 0~500N), used for collecting joint motion parameters and interaction force signal.

[0051] (2) Biological signal acquisition equipment: (1) Surface electromyography sensor: An 8-channel surface electromyography acquisition system is used with a sampling frequency of 2000Hz. The electrodes are placed on the muscle bellies of the rectus femoris, biceps femoris, tibialis anterior, and gastrocnemius muscles in the lower extremities of the human body to collect surface electromyography signals.

[0052] (2) Acceleration sensor: A three-axis acceleration sensor is used with a sampling frequency of 1000Hz. It is placed on the waist and ankle of the human body to collect human motion acceleration signals.

[0053] (3) Control module: An embedded controller (such as STM32H7 series microcontroller) is used, equipped with a high-speed AD acquisition module and a PWM output module, to realize the acquisition, processing, motion intention recognition, control parameter adjustment, and control instruction generation and transmission of multi-source biological signals.

[0054] (4) Software algorithm: MATLAB and C language are used to develop related algorithm programs, including multi-source biological signal preprocessing algorithm, CNN-RNN motion intention classification algorithm based on attention mechanism, and fuzzy PID control parameter adjustment algorithm.

[0055] II. Experimental process (I) Establishment of individual motion characteristic model of user: (1) Select 10 healthy volunteers (age 20-30 years old, height 165-185 cm, weight 55-80 kg) as experimental objects. Each volunteer performs a 5-minute warm-up exercise before the experiment.

[0056] (2) Control the exoskeleton robot in passive following mode, guide the volunteers to complete three standard motion modes: standard walking (step speed 0.8-1.2m / s), standard running (step speed 2.5-3.5m / s), and going up and down stairs (step height 15cm), each lasting 30 seconds. At the same time, collect the multi-source biological signals of the volunteers and the motion parameters of the exoskeleton robot.

[0057] (3) Preprocess and extract features from the collected multi-source biological signals, combine with the exoskeleton motion parameters, and establish individual motion characteristic models of each volunteer under three standard motion modes, stored in the database of the control module.

[0058] (II) Real-time collaborative control experiment: (1) The volunteer wears the exoskeleton robot, and the control module calls the individual motion characteristic model of the volunteer from the database to initialize the collaborative control parameters.

[0059] (2) Guide the volunteers to walk, run, go up and down the stairs and other movements according to their own will. During the movement, the control module collects multi-source biological signals in real time, pre-processes and extracts features, inputs the CNN-RNN model based on the attention mechanism, and identifies the human movement intention.

[0060] (3) According to the identified movement intention, the control module compares the current biological signal features with the reference features in the individual movement feature model, combines the interaction force signals between the exoskeleton and the human body, evaluates the real-time movement state and fatigue degree of the volunteer, and uses the fuzzy PID control algorithm to adaptively adjust the cooperative control parameters.

[0061] (4) The control module generates control instructions for each joint of the exoskeleton according to the movement intention and the adjusted control parameters, drives the exoskeleton to move, simultaneously receives the movement feedback signals of the exoskeleton in real time, calculates the movement error and corrects the control instructions.

[0062] (Three) Experimental data recording and analysis: (1) During the experiment, the following data are recorded in real time: human movement intention recognition accuracy, angle deviation of exoskeleton robot and human joint movement, cooperative control response time, and subjective comfort score of volunteers (using 1-10 point system, 1 point for extremely uncomfortable, 10 points for very comfortable).

[0063] (2) Each volunteer repeats the experiment 3 times under each movement mode, and the average value is taken as the final experimental result.

[0064] Three, experimental results and analysis (1) Movement intention recognition accuracy: the experimental results show that under the three movement modes of walking, running and going up and down the stairs, the human movement intention recognition accuracy reaches 96.5%, 95.2% and 94.8% respectively, and the average recognition accuracy is 95.5%. Compared with the existing single signal recognition method (average accuracy about 85%), the recognition accuracy is significantly improved, which shows that the multi-source signal fusion and attention mechanism deep learning model of the invention can effectively improve the accuracy of movement intention recognition.

[0065] (2) Joint movement angle deviation: the angle deviation of exoskeleton robot and human joint movement is 0.8° in walking mode, 1.2° in running mode, and 1.5° in going up and down the stairs mode, which is less than the preset error threshold (2°), indicating that the cooperative control method of the invention can realize precise cooperation of exoskeleton and human joint movement, and the movement error is small.

[0066] (3) Synergy control response time: The synergy control response time (the time from the change of the human body movement state to the adjustment of the exoskeleton) is 50 ms on average in the three movement modes, which is 50% higher than the existing control method (the average response time is about 100 ms), effectively solving the problem of synergy control response lag.

[0067] (4) Subjective comfort score: The subjective comfort score of the volunteers in the three movement modes is 8.5 on average, of which the walking mode is 9.0 on average, the running mode is 8.3 on average, and the up and down stairs mode is 8.2 on average, indicating that the volunteers feel more comfortable when using the synergy control method of the application to control the exoskeleton robot, which reduces the physical burden and has a good user experience.

[0068] In summary, the exoskeleton robot and the human limb movement synergy control method can effectively improve the movement intention recognition accuracy, reduce the synergy control response lag, enhance the synergy control adaptability, and at the same time improve the user experience and safety, which has high practical value and popularization prospect.

[0069] Figure 2 A module block diagram of an exoskeleton robot synergy control device is provided for the embodiment of the application.

[0070] Optionally, as another embodiment of the application, as shown in Figure 2 An exoskeleton robot synergy control device comprises: A first signal obtaining module is used to obtain a plurality of original surface electromyography signals, original joint angle signals corresponding to each original surface electromyography signal, original joint angular velocity signals corresponding to each original surface electromyography signal, and original human body movement acceleration signals corresponding to each original surface electromyography signal through a sensor group arranged at a preset part of the human body. A preprocessing module is used to preprocess each original surface electromyography signal, the original joint angle signal corresponding to each original surface electromyography signal, the original joint angular velocity signal corresponding to each original surface electromyography signal, and the original human body movement acceleration signal corresponding to each original surface electromyography signal, respectively, to obtain a preprocessed surface electromyography signal corresponding to each original surface electromyography signal, a preprocessed joint angle signal corresponding to each original surface electromyography signal, a preprocessed joint angular velocity signal corresponding to each original surface electromyography signal, and a preprocessed human body movement acceleration signal corresponding to each original surface electromyography signal. a feature extraction module configured to extract original time domain features corresponding to each of the original sEMG signals, original joint angle features corresponding to each of the original sEMG signals, original joint angular velocity features corresponding to each of the original sEMG signals, and original acceleration features corresponding to each of the original sEMG signals from each of the preprocessed sEMG signals, preprocessed joint angle signals corresponding to each of the original sEMG signals, preprocessed joint angular velocity signals corresponding to each of the original sEMG signals, and preprocessed human motion acceleration signals corresponding to each of the original sEMG signals, respectively; a fusion analysis module configured to construct a classification model, and perform fusion analysis on all of the original time domain features, all of the original joint angle features, all of the original joint angular velocity features, and all of the original acceleration features through the classification model to obtain human motion intention category information; a second signal obtaining module configured to obtain a plurality of interaction force signals through force sensors arranged on the exoskeleton robot; a parameter analysis module configured to perform parameter analysis on all of the original time domain features, all of the original joint angle features, all of the original joint angular velocity features, all of the original acceleration features, and all of the interaction force signals to obtain target exoskeleton robot control parameters; a control module configured to generate motion control instructions according to the human motion intention category information and the target exoskeleton robot control parameters, and control the exoskeleton robot to move according to the motion control instructions.

[0071] Optionally, as an embodiment of the present application, the preprocessing module is specifically configured to: filter each of the original sEMG signals to obtain filtered sEMG signals corresponding to each of the original sEMG signals; rectify each of the filtered sEMG signals to obtain rectified sEMG signals corresponding to each of the original sEMG signals; integrate each of the rectified sEMG signals by using a sliding window integration algorithm to obtain preprocessed sEMG signals corresponding to each of the original sEMG signals; smooth each of the original joint angle signals and the original joint angular velocity signals corresponding to each of the original sEMG signals to obtain smoothed joint angle signals corresponding to each of the original sEMG signals and smoothed joint angular velocity signals corresponding to each of the original sEMG signals, respectively; filtering processing is performed on each of the smoothed joint angle signals and the smoothed joint angular velocity signals corresponding to each of the original surface electromyography signals, to obtain preprocessed joint angle signals corresponding to each of the original surface electromyography signals and preprocessed joint angular velocity signals corresponding to each of the original surface electromyography signals; calibration processing is performed on each of the original human motion acceleration signals, to obtain calibrated human motion acceleration signals corresponding to each of the original surface electromyography signals; filtering processing is performed on each of the calibrated human motion acceleration signals by using a low-pass filter, to obtain preprocessed human motion acceleration signals corresponding to each of the original surface electromyography signals.

[0072] Optionally, as an embodiment of the present application, the classification model comprises a convolutional neural network, a recurrent neural network, an attention mechanism layer and a fully connected layer; In the fusion analysis module, the process of performing fusion analysis on all the original time domain features, all the original joint angle features, all the original joint angular velocity features and all the original acceleration features by using the classification model to obtain human motion intention category information comprises: local feature extraction is performed on each of the original time domain features, the original joint angle features corresponding to each of the original surface electromyography signals, the original joint angular velocity features corresponding to each of the original surface electromyography signals and the original acceleration features corresponding to each of the original surface electromyography signals by using the convolutional neural network, to obtain local time domain features corresponding to each of the original surface electromyography signals, local joint angle features corresponding to each of the original surface electromyography signals, local joint angular velocity features corresponding to each of the original surface electromyography signals and local acceleration features corresponding to each of the original surface electromyography signals; dynamic feature extraction is performed on each of the local time domain features, the local joint angle features corresponding to each of the original surface electromyography signals, the local joint angular velocity features corresponding to each of the original surface electromyography signals and the local acceleration features corresponding to each of the original surface electromyography signals by using the recurrent neural network, to obtain dynamic time domain features corresponding to each of the original surface electromyography signals, dynamic joint angle features corresponding to each of the original surface electromyography signals, dynamic joint angular velocity features corresponding to each of the original surface electromyography signals and dynamic acceleration features corresponding to each of the original surface electromyography signals; fusion is performed on all the dynamic time domain features, all the dynamic joint angle features, all the dynamic joint angular velocity features and all the dynamic acceleration features by using the attention mechanism layer, to obtain target fusion features; The target fusion feature is classified by the full connection layer to obtain human motion intention category information.

[0073] Optionally, another embodiment of the present application provides an exoskeleton robot cooperative control system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the exoskeleton robot cooperative control method as described above is realized. The system can be a computer or the like.

[0074] Optionally, another embodiment of the present application provides a computer readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the exoskeleton robot cooperative control method as described above is realized.

[0075] It should be noted that, in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0076] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0077] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0078] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment of the present application.

[0079] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0080] If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in the form of a contribution to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0081] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A collaborative control method for an exoskeleton robot, characterized in that, Includes the following steps: Multiple raw surface electromyography (EMG) signals, raw joint angle signals corresponding to each raw EMG signal, raw joint angular velocity signals corresponding to each raw EMG signal, and raw human motion acceleration signals corresponding to each raw EMG signal are obtained by a sensor group set at a preset part of the human body. Each of the original surface electromyography (SEMG) signals, the original joint angle signals corresponding to each of the original SEMG signals, the original joint angular velocity signals corresponding to each of the original SEMG signals, and the original human motion acceleration signals corresponding to each of the original SEMG signals are preprocessed to obtain preprocessed SEMG signals, preprocessed joint angle signals, preprocessed joint angular velocity signals, and preprocessed human motion acceleration signals corresponding to each of the original SEMG signals. The original time-domain features, original joint angle features, original joint angular velocity features, and original acceleration features corresponding to each original surface electromyography (EMG) signal are extracted from each of the preprocessed EMG signals, the preprocessed joint angle signals corresponding to each original EMG signals, the preprocessed joint angular velocity signals corresponding to each original EMG signals, and the preprocessed human motion acceleration signals corresponding to each original EMG signals, respectively. A classification model is constructed, and the original time-domain features, original joint angle features, original joint angular velocity features, and original acceleration features are fused and analyzed through the classification model to obtain human motion intention category information. Multiple interactive force signals are obtained by force sensors installed on the exoskeleton robot; Parametric analysis is performed on all the original time-domain features, all the original joint angle features, all the original joint angular velocity features, all the original acceleration features, and all the interactive force signals to obtain the control parameters of the target exoskeleton robot. Based on the human motion intention category information and the target exoskeleton robot control parameters, motion control commands are generated, and the exoskeleton robot is controlled to move according to the motion control commands.

2. The exoskeleton robot collaborative control method according to claim 1, characterized in that, The process of preprocessing each of the original surface electromyography (EMG) signals, the corresponding original joint angle signals, the corresponding original joint angular velocity signals, and the corresponding original human motion acceleration signals to obtain preprocessed EMG signals, preprocessed joint angle signals, preprocessed joint angular velocity signals, and preprocessed human motion acceleration signals corresponding to each of the original EMG signals includes: Each of the original surface electromyography (EMG) signals is filtered to obtain a filtered surface EMG signal corresponding to each of the original surface EMG signals. Each of the filtered surface electromyography (EMG) signals is rectified to obtain rectified surface EMG signals corresponding to each of the original surface EMG signals. The sliding window integration algorithm is used to integrate each of the rectified surface electromyography (EMG) signals to obtain the preprocessed surface EMG signals corresponding to each of the original surface EMG signals. Each of the original joint angle signals and the original joint angular velocity signals corresponding to each of the original surface electromyography signals are smoothed to obtain smoothed joint angle signals and smoothed joint angular velocity signals corresponding to each of the original surface electromyography signals. Each smoothed post-joint angle signal and the smoothed post-joint angular velocity signal corresponding to each original surface electromyography signal are filtered to obtain pre-processed post-joint angle signals and pre-processed post-joint angular velocity signals corresponding to each original surface electromyography signal. Each of the original human motion acceleration signals is calibrated to obtain calibrated human motion acceleration signals corresponding to each of the original surface electromyography signals. Each of the calibrated human motion acceleration signals is filtered using a low-pass filter to obtain a preprocessed human motion acceleration signal corresponding to each of the original surface electromyography signals.

3. The exoskeleton robot collaborative control method according to claim 1, characterized in that, The classification model includes a convolutional neural network, a recurrent neural network, an attention mechanism layer, and a fully connected layer; The process of fusing and analyzing all the original time-domain features, all the original joint angle features, all the original joint angular velocity features, and all the original acceleration features through the classification model to obtain human motion intention category information includes: The convolutional neural network extracts local features from each of the original temporal features, the original joint angle features corresponding to each of the original surface electromyography (EMG) signals, the original joint angular velocity features corresponding to each of the original EMG signals, and the original acceleration features corresponding to each of the original EMG signals, thereby obtaining local temporal features, local joint angle features, local joint angular velocity features, and local acceleration features corresponding to each of the original EMG signals. The recurrent neural network is used to extract dynamic features from each of the local temporal features, the local joint angle features corresponding to each of the original surface electromyography (EMG) signals, the local joint angular velocity features corresponding to each of the original EMG signals, and the local acceleration features corresponding to each of the original EMG signals, to obtain the dynamic temporal features, dynamic joint angle features, dynamic joint angular velocity features, and dynamic acceleration features corresponding to each of the original EMG signals. The attention mechanism layer fuses all the dynamic temporal features, all the dynamic joint angle features, all the dynamic joint angular velocity features, and all the dynamic acceleration features to obtain the target fusion feature. The target fusion features are classified by the fully connected layer to obtain human motion intention category information.

4. The exoskeleton robot collaborative control method according to claim 1, characterized in that, The process of performing parameter analysis on all the original time-domain features, all the original joint angle features, all the original joint angular velocity features, all the original acceleration features, and all the interaction force signals to obtain the control parameters of the target exoskeleton robot includes: The difference between each of the original time-domain features and the standard time-domain features in the pre-constructed individual motion feature library is calculated to obtain the time-domain feature deviation value corresponding to each of the original surface electromyography signals; The difference between each of the original joint angle features and the standard joint angle features in the pre-constructed individual motion feature library is calculated to obtain the joint angle feature deviation value corresponding to each of the original surface electromyography signals; The difference between each of the original joint angular velocity features and the standard joint angular velocity features in the pre-constructed individual motion feature library is calculated to obtain the joint angular velocity feature deviation value corresponding to each of the original surface electromyography signals; The difference between each of the original acceleration features and the standard acceleration features in the pre-constructed individual motion feature library is calculated to obtain the acceleration feature deviation value corresponding to each of the original surface electromyography signals; Multiple standard acceleration values ​​are extracted from a pre-built database of individual motion features; Stability assessment analysis is performed on all the original acceleration characteristics, all the interactive force signals, and all the standard acceleration values ​​to obtain stability assessment results; The initial exoskeleton robot control parameters are imported, and then adjusted according to the fuzzy PID control algorithm, the stability evaluation results, all time-domain characteristic deviation values, all joint angle characteristic deviation values, all joint angular velocity characteristic deviation values, and all acceleration characteristic deviation values ​​to obtain the target exoskeleton robot control parameters.

5. The exoskeleton robot collaborative control method according to claim 4, characterized in that, The process of performing stability assessment analysis on all the original acceleration characteristics, all the interaction force signals, and all the standard acceleration values ​​to obtain the stability assessment results includes: The interaction force amplitude corresponding to each of the original surface electromyography signals is extracted from each of the interaction force signals; Calculate the standard deviation of all the said interaction force amplitudes to obtain the standard deviation of the interaction force amplitude; Calculate the average value of all the interaction force amplitudes to obtain the average interaction force amplitude; The interaction force fluctuation coefficient is obtained by calculating the standard deviation of the interaction force amplitude and the average value of the interaction force amplitude using the first formula, which is: ， in, For interaction force fluctuation coefficient, The standard deviation of the interaction force amplitude. This represents the average value of the interaction force amplitude; The original acceleration values ​​corresponding to the original surface electromyography signals are extracted from each of the original acceleration features; Calculate the standard deviation of all the original acceleration values ​​to obtain the original acceleration standard deviation; The acceleration trajectory deviation value is obtained by calculating all the original acceleration values ​​and all the standard acceleration values ​​using the second equation, which is: ， in, This represents the acceleration trajectory deviation value. The number of original acceleration values. For the first One original acceleration value, For the first A standard acceleration value; The stability score is obtained by calculating the interaction force fluctuation coefficient, the original acceleration standard deviation, and the acceleration trajectory deviation value using the third equation: ， in, For stability rating, For interaction force fluctuation coefficient, The original standard deviation of acceleration. This represents the acceleration trajectory deviation value. If the stability score is greater than or equal to a preset first threshold, then the first preset evaluation result is taken as the stability evaluation result. If the stability score is less than the preset first threshold and greater than or equal to the preset second threshold, then the second preset evaluation result shall be used as the stability evaluation result. If the stability score is less than the preset second threshold, then the third preset evaluation result will be used as the stability evaluation result.

6. A collaborative control device for an exoskeleton robot, characterized in that, include: The first signal acquisition module is used to acquire multiple raw surface electromyography signals, raw joint angle signals corresponding to each raw surface electromyography signal, raw joint angular velocity signals corresponding to each raw surface electromyography signal, and raw human motion acceleration signals corresponding to each raw surface electromyography signal through a sensor group set at a preset part of the human body. The preprocessing module is used to preprocess each of the original surface electromyography (SEMG) signals, the original joint angle signals corresponding to each of the original SEMG signals, the original joint angular velocity signals corresponding to each of the original SEMG signals, and the original human motion acceleration signals corresponding to each of the original SEMG signals, to obtain preprocessed SEMG signals, preprocessed joint angle signals, preprocessed joint angular velocity signals, and preprocessed human motion acceleration signals corresponding to each of the original SEMG signals. The feature extraction module is used to extract, respectively, the original time-domain features corresponding to each original surface electromyography (SEMG) signal, the original joint angle features corresponding to each original SEMG signal, the original joint angular velocity features corresponding to each original SEMG signal, and the original acceleration features corresponding to each original SEMG signal from each of the preprocessed SEMG signals, the preprocessed joint angle signals corresponding to each original SEMG signals, the preprocessed joint angular velocity signals corresponding to each original SEMG signals, and the preprocessed human motion acceleration signals corresponding to each original SEMG signals; The fusion analysis module is used to construct a classification model, and through the classification model, to perform fusion analysis on all the original time-domain features, all the original joint angle features, all the original joint angular velocity features, and all the original acceleration features to obtain human motion intention category information. The second signal acquisition module is used to acquire multiple interactive force signals through force sensors installed on the exoskeleton robot; The parameter analysis module is used to perform parameter analysis on all the original time-domain features, all the original joint angle features, all the original joint angular velocity features, all the original acceleration features, and all the interaction force signals to obtain the control parameters of the target exoskeleton robot. The control module is used to generate motion control commands based on the human motion intention category information and the target exoskeleton robot control parameters, and to control the exoskeleton robot to move according to the motion control commands.

7. The exoskeleton robot collaborative control device according to claim 6, characterized in that, The preprocessing module is specifically used for: Each of the original surface electromyography (EMG) signals is filtered to obtain a filtered surface EMG signal corresponding to each of the original surface EMG signals. Each of the filtered surface electromyography (EMG) signals is rectified to obtain rectified surface EMG signals corresponding to each of the original surface EMG signals. The sliding window integration algorithm is used to integrate each of the rectified surface electromyography (EMG) signals to obtain the preprocessed surface EMG signals corresponding to each of the original surface EMG signals. Each of the original joint angle signals and the original joint angular velocity signals corresponding to each of the original surface electromyography signals are smoothed to obtain smoothed joint angle signals and smoothed joint angular velocity signals corresponding to each of the original surface electromyography signals. Each smoothed post-joint angle signal and the smoothed post-joint angular velocity signal corresponding to each original surface electromyography signal are filtered to obtain pre-processed post-joint angle signals and pre-processed post-joint angular velocity signals corresponding to each original surface electromyography signal. Each of the original human motion acceleration signals is calibrated to obtain calibrated human motion acceleration signals corresponding to each of the original surface electromyography signals. Each of the calibrated human motion acceleration signals is filtered using a low-pass filter to obtain a preprocessed human motion acceleration signal corresponding to each of the original surface electromyography signals.

8. The exoskeleton robot collaborative control device according to claim 6, characterized in that, The classification model includes a convolutional neural network, a recurrent neural network, an attention mechanism layer, and a fully connected layer; In the fusion analysis module, the process of fusing and analyzing all the original time-domain features, all the original joint angle features, all the original joint angular velocity features, and all the original acceleration features through the classification model to obtain human motion intention category information includes: The convolutional neural network extracts local features from each of the original temporal features, the original joint angle features corresponding to each of the original surface electromyography (EMG) signals, the original joint angular velocity features corresponding to each of the original EMG signals, and the original acceleration features corresponding to each of the original EMG signals, thereby obtaining local temporal features, local joint angle features, local joint angular velocity features, and local acceleration features corresponding to each of the original EMG signals. The recurrent neural network is used to extract dynamic features from each of the local temporal features, the local joint angle features corresponding to each of the original surface electromyography (EMG) signals, the local joint angular velocity features corresponding to each of the original EMG signals, and the local acceleration features corresponding to each of the original EMG signals, to obtain the dynamic temporal features, dynamic joint angle features, dynamic joint angular velocity features, and dynamic acceleration features corresponding to each of the original EMG signals. The attention mechanism layer fuses all the dynamic temporal features, all the dynamic joint angle features, all the dynamic joint angular velocity features, and all the dynamic acceleration features to obtain the target fusion feature. The target fusion features are classified by the fully connected layer to obtain human motion intention category information.

9. A collaborative control device for an exoskeleton robot, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the exoskeleton robot collaborative control method as described in any one of claims 1 to 5.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the exoskeleton robot collaborative control method as described in any one of claims 1 to 5.

Images