Exoskeleton manipulator control method

Abstract

The invention discloses an exoskeleton manipulator control method which comprises the following steps: constructing a hand biomechanical model, defining the degree of freedom of finger joints, and carrying out joint division modeling based on a drive kinetic equation of a Bowden cable; deploying a high-density sensor on the hand; s3, a hand specificity Bouc-Wen model is improved, an inter-finger coupling term is introduced to correct a lag state equation, and micro-lag parameters are identified through a hierarchical identification algorithm; s4, designing a self-adaptive cooperative control strategy, realizing fuzzy mapping from the electromyographic signals to the joint torque, and optimizing a cooperative tracking error through a multi-joint parameter coupling self-adaptive law; s5, dynamically compensating the mechanical resistance of the hand by adopting the corrected double-Kalman filter; and S6, optimizing closed-loop execution and a security policy. According to the exoskeleton manipulator control method, through the improved Bouc-Wen model and the self-adaptive cooperative control strategy, the mapping relation between the electromyographic signal and the action of the manipulator can be effectively processed, and the exoskeleton manipulator control method adapts to different operation environments.

Description

Technical Field

[0001] The invention belongs to the technical field of manipulator control, and in particular relates to an exoskeleton manipulator control method. Background Art

[0002] Exoskeleton manipulators are wearable devices that can enhance the functions of human hands and are widely used in rehabilitation medicine, industrial operations and military fields. Existing exoskeleton manipulator control methods have the following problems: the sensor configuration is not perfect, and the contact force and joint angle information of the hand cannot be obtained in real time; the hysteresis model fails to fully consider the multi-joint coupling effect, resulting in large control errors; and there is a lack of adaptive collaborative control strategy, which cannot effectively process the mapping relationship between electromyographic signals and manipulator movements. Summary of the invention

[0003] The present invention provides an exoskeleton manipulator control method to solve the above-mentioned technical problems, and specifically adopts the following technical solutions: A method for controlling an exoskeleton manipulator comprises the following steps: S1: Construct a hand biomechanical model, define the degrees of freedom of the finger joints, and perform joint modeling based on the driving dynamics equations of the Bowden cable; S2: Deploy high-density sensors on the hand, including fingertip tactile arrays, joint micro-encoders, and surface electromyography sensors. The fingertip tactile array is used to measure contact force distribution, the joint micro-encoders are used to measure joint angles, and the surface electromyography sensors are used to collect electromyographic signals of the forearm flexor / extensor muscle groups. S3: Improve the hand-specific Bouc-Wen model, introduce inter-finger coupling terms to correct the hysteresis state equation, and identify the micro-hysteresis parameters through a hierarchical identification algorithm; S4: Design an adaptive collaborative control strategy to realize the fuzzy mapping from EMG signals to joint torques, and optimize the collaborative tracking error through the multi-joint parameter coupling adaptive law; S5: A modified dual Kalman filter is used to dynamically compensate for the mechanical resistance of the hand, including a state filter and a parameter filter, which are used to estimate the joint friction force and update the friction coefficient in real time, respectively; S6: Optimize closed-loop execution and safety strategies, including variable impedance control, finger collaborative trajectory optimization, and safety boundary protection mechanism.

[0004] Furthermore, in step S1, the finger joints have 14 degrees of freedom, and the dynamic equation of each joint is:

[0005] Where i=1,2,…,14 represents the joint number, τ i is the driving torque of the i-th joint, Ji is the inertia term, θ i is the joint angle, B i is the damping term, K i is the elastic term, H i (z i ) is the lag term, τ ext,i is the external torque term, F fric,i is the i-th joint friction term, which includes the coupling effect of fingertip contact friction and joint bearing friction.

[0006] Furthermore, in step S2, the fingertip tactile array is a 4*4 piezoresistive sensor, and the calculation formula for the contact force distribution is:

[0007] Among them, w j is the normalized weight, P j is the measurement value of the jth sensor, w j ∈[0,1],∑w j =1.

[0008] Further, in step S3, the improved Bouc-Wen hysteresis state equation is:

[0009] Among them, z i represents the hysteresis state variable of the i-th joint, C ij is the inter-finger coupling coefficient, calibrated through grasping experiments; The micro-hysteresis parameters are identified by a hierarchical identification algorithm, and the outer layer uses the PSO optimization algorithm to optimize α i , β i , γ i and n,α i is the hysteresis stiffness coefficient of the i-th joint, controlling the hysteresis linear part, β i is the nonlinear hysteresis damping parameter of the i-th joint, affecting the saturation rate, γ i is the nonlinear hysteresis shape parameter of the i-th joint, n∈[1,3], is the hysteresis nonlinear order; The inner layer uses the least squares method to update C in real time ij and F fric,i .

[0010] Furthermore, in step S4, the fuzzy mapping formula from the electromyographic signal to the joint torque is:

[0011] Among them, a m is the myoelectric activation degree of the mth channel, a th is the individual activation threshold, K sEMG=0.5N*m / V, represents the electromyographic signal-torque conversion gain coefficient, which is determined by calibration experiments, θ max,i Represents the maximum allowed angle of the i-th joint, avoiding sudden changes in the mapping value when the joint angle approaches the limit.

[0012] Further, in step S4, the collaborative tracking error is defined as:

[0013] Where λ=0.3, w i is the weight coefficient, w i ∝1 / J i , N=14, θ i is the angle of the i-th joint, θ j is the angle of the jth joint; The improved adaptive algorithm is:

[0014] in, represents the joint equivalent inertia adjusted in real time, where ^ represents the derivative, γJ is the adaptive gain, which is used to control the update rate of the inertia parameters, and e i is the tracking error of the i-th joint, J nom is the nominal inertia value, which is set according to the initial biomechanical model. max It is the maximum inertia value to prevent parameter overshoot.

[0015] Furthermore, in step S5, the prediction and update equations of the state filter are:

[0016] Among them A i = diag[0.92, 0.85],A i is the state transfer matrix, describing the dynamic attenuation characteristics of the friction state, calibrated by step response experiments. represents the first derivative of the angle of the i-th joint, represents the first derivative of the angle of the i-th joint at the previous moment, is the friction state vector of the i-th joint, represents the friction state vector of the i-th joint at the previous moment, K f,i represents the Kalman gain matrix of the i-th joint state filter, F tip,i represents the contact force distribution of the i-th fingertip tactile array, , H i is the observation matrix, which maps the friction state to the sensor measurement space, where =0.01 rad / s, is the velocity normalization factor, B fric,i is the input coupling matrix, which is the excitation weight of the angular velocity on the friction state.

[0017] Furthermore, in step S5, the update equation of the parameter filter is:

[0018] Among them, σ i is the friction coefficient, is the friction coefficient at the current moment, represents the friction coefficient at the last moment, K σ,i is the parameter update gain, which controls the parameter convergence rate. is the model sensitivity matrix; The compensation torque formula for fingertip contact force feedforward compensation is:

[0019] Among them, τ comp,i is the compensation torque of the ith joint, x j is the coordinate of the fingertip contact point, F tip,j represents the contact force of the jth sensor in the 4*4 fingertip tactile array, represents the derivative of the i-th joint angle to the j-th sensor in each fingertip tactile array, K tip Represents the mapping coefficient, which is used to gain the contact force mapping.

[0020] Further, step S6 is specifically as follows: The impedance parameters are adjusted according to the stiffness of the grasped object to achieve variable impedance control, and it automatically switches to the admittance control mode when a sudden external force is detected; The finger collaborative trajectory is corrected online based on the contact force distribution to ensure the stability of multi-finger envelope grasping; A safety boundary protection mechanism is set up to ensure the safe operation of the exoskeleton manipulator through joint angle-torque envelope constraints.

[0021] Furthermore, in step S6, the impedance parameter adjustment formula of the variable impedance control is:

[0022] Among them, K p is the stiffness parameter, K p0 is the set foundation stiffness parameter, is the stiffness variation coefficient, B d is the damping parameter, B d0 is the basic damping parameter set, Indicates F tip The second-order vector norm of ; The online correction formula of finger collaborative trajectory is:

[0023] in, is the corrected real-time reference angle of the i-th joint, θ d,i is the original expected angle of the i-th joint, which comes from the intention trajectory generated by the electromyographic signal, neighbors is the set of collaborative joints, including the joints of all fingers involved in the enveloping grasp, θ d,j is the expected angle of the neighboring joint j associated with joint i, H(z i ) is used to convert the delayed contact state z i Converted into equivalent potential energy, K c is the synergistic gain, K h is the lag term gain; The joint angle-torque envelope constraint formula of the safety boundary protection mechanism is:

[0024] Among them, τ cmd,i is the final output control torque of the i-th joint, τ comp,i is the compensation torque of the i-th joint, τ i is the actual torque output at present, τ max is the maximum output torque limit, τ soft It is the buffer target torque when the joint approaches the safety limit, which reduces the joint impact by reducing the output torque, η=sigmoid(|θ i -θ max,i ∣) to achieve a smooth transition.

[0025] The invention is beneficial in that the exoskeleton manipulator control method provided by the invention can obtain the contact force and joint angle information of the hand in real time through the accurate hand biomechanics model and high-density sensor configuration, thereby improving the control accuracy. Through the improved Bouc-Wen model and adaptive collaborative control strategy, the mapping relationship between the electromyographic signal and the manipulator action can be effectively processed to adapt to different operating environments.

[0026] The invention is also beneficial in that the exoskeleton manipulator control method provided uses a modified dual Kalman filter to dynamically compensate for the mechanical resistance of the hand, which can estimate and compensate for the joint friction in real time and improve the stability of control. Through variable impedance control and safety boundary protection mechanism, the impedance parameters can be dynamically adjusted according to the stiffness of the grasped object, and when a sudden external force is detected, it can automatically switch to the admittance control mode to ensure the safe operation of the exoskeleton manipulator. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0028] Figure 1 It is a schematic diagram of the exoskeleton manipulator control method of the present application. DETAILED DESCRIPTION

[0029] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0030] like Figure 1 The present invention shows an exoskeleton manipulator control method, which includes the following steps: S1: Construct a hand biomechanical model, define the degrees of freedom of the finger joints, and perform joint modeling based on the driving dynamics equations of the Bowden cable.

[0031] S2: Deploy high-density sensors on the hand, including fingertip tactile arrays, joint micro-encoders, and surface electromyography sensors. The fingertip tactile array is used to measure contact force distribution, the joint micro-encoders are used to measure joint angles, and the surface electromyography sensors are used to collect electromyographic signals of the forearm flexor / extensor muscles. The surface electromyography sensors use 8-channel differential motors and are deployed in the forearm flexor / extensor muscles.

[0032] S3: The hand-specific Bouc-Wen model is improved, the inter-finger coupling term is introduced to correct the hysteresis state equation, and the micro-hysteresis parameters are identified through a hierarchical identification algorithm.

[0033] S4: Design an adaptive collaborative control strategy to realize the fuzzy mapping of EMG signals to joint torques, and optimize the collaborative tracking error through the multi-joint parameter coupling adaptive law.

[0034] S5: A modified dual Kalman filter is used to dynamically compensate for the mechanical resistance of the hand, including a state filter and a parameter filter, which are used to estimate the joint friction force and update the friction coefficient in real time, respectively.

[0035] S6: Optimize closed-loop execution and safety strategies, including variable impedance control, finger collaborative trajectory optimization, and safety boundary protection mechanism.

[0036] In the implementation of the present application, in step S1, the finger joints have 14 degrees of freedom, and the dynamic equation of each joint is:

[0037] Where i=1,2,…,14 represents the joint number, τ i is the driving torque of the i-th joint, J i is the inertia term, θ i is the joint angle, B i is the damping term, K i is the elastic term, H i (z i ) is the lag term, τ ext,i is the external torque term, F fric,i is the i-th joint friction term, which includes the coupling effect of fingertip contact friction and joint bearing friction.

[0038] In the embodiment of the present application, in step S2, the fingertip tactile array is a 4*4 piezoresistive sensor with a sampling rate of 1kHz, and the calculation formula of the contact force distribution is:

[0039] Among them, w j is the normalized weight, P j For the j The measured value of the sensor, w j ∈[0,1],∑w j =1.

[0040] In the implementation manner of the present application, in step S3, the fingertip coupling term is introduced, and the improved Bouc-Wen hysteresis state equation is:

[0041] Among them, z i represents the hysteresis state variable of the i-th joint, C ij is the inter-finger coupling coefficient, which is calibrated through grasping experiments.

[0042] The finger micro-motion excitation signal is designed to have an amplitude of ±5° and a frequency of 0.5-10Hz. The micro-hysteresis parameters are identified by a hierarchical identification algorithm, and the outer layer uses a PSO optimization algorithm to optimize α i , β i , γ i and n,α i is the hysteresis stiffness coefficient of the i-th joint, controlling the hysteresis linear part, β i is the nonlinear hysteresis damping parameter of the i-th joint, affecting the saturation rate, γ i is the nonlinear hysteresis shape parameter of the i-th joint, n∈[1,3], and is the hysteresis nonlinear order.

[0043] The inner layer uses the least squares method to update C in real time ij and F fric,i .

[0044] In the implementation of the present application, in step S4, the fuzzy mapping formula from the electromyographic signal to the joint torque is:

[0045] Among them, a m is the electromyographic activation of the mth channel (RMS calculation window 10ms), a th is the individual activation threshold (adaptively calibrated when used for the first time), K sEMG =0.5N*m / V, represents the electromyographic signal-torque conversion gain coefficient, which is determined by calibration experiments, θ max,i Represents the maximum allowed angle of the i-th joint, avoiding sudden changes in the mapping value when the joint angle approaches the limit.

[0046] In the implementation of the present application, in step S4, the collaborative tracking error is defined as:

[0047] Where λ=0.3, w i is the weight coefficient, w i ∝1 / J i , N=14, θ i is the angle of the i-th joint, θ j is the angle of the jth joint.

[0048] The improved adaptive algorithm is:

[0049] in, represents the joint equivalent inertia adjusted in real time, where ^ represents the derivative, γJ is the adaptive gain, which is used to control the update rate of the inertia parameters, and e i is the tracking error of the i-th joint, J nom is the nominal inertia value, which is set according to the initial biomechanical model. max is the maximum inertia value.

[0050] In the implementation manner of the present application, in step S5, the prediction and update equations of the state filter are:

[0051] Among them A i = diag[0.92, 0.85] (including static friction and dynamic friction states), A iis the state transfer matrix, describing the dynamic attenuation characteristics of the friction state, calibrated by step response experiments. represents the first derivative of the angle of the i-th joint, represents the first derivative of the angle of the i-th joint at the previous moment, is the friction state vector of the i-th joint (positive and negative are the dynamic and static friction components, respectively), where the superscript ^ indicates that it is processed at the derivative level. represents the friction state vector of the i-th joint at the previous moment, K f,i represents the Kalman gain matrix of the i-th joint state filter, F tip,i represents the contact force distribution of the i-th fingertip tactile array, , H i is the observation matrix, which maps the friction state to the sensor measurement space, where =0.01 rad / s, is the velocity normalization factor, B fric,i is the input coupling matrix, which is the excitation weight of the angular velocity on the friction state. In the prediction equation, based on the last moment (i.e. k-1 ) is used to derive the predicted friction state vector from the first derivative of the angle of the i-th joint and the friction state vector. Then, in the update equation, the predicted friction state vector and the force measurement value F of the i-th fingertip are used to tip,i An updated predicted friction state vector is obtained to achieve state filtering.

[0052] In the implementation manner of the present application, in step S5, the update equation of the parameter filter is:

[0053] Among them, σ i is the friction coefficient, is the friction coefficient at the current moment, represents the friction coefficient at the last moment, K σ,i is the parameter update gain, which controls the parameter convergence rate. is the model sensitivity matrix.

[0054] The constraints are: 0,i > 0.1N, σ 1,i ∈ [0.01,0.1] N·s / rad.

[0055] The compensation torque formula for fingertip contact force feedforward compensation is:

[0056] Among them, τ comp,i is the compensation torque of the ith joint, x j is the coordinate of the fingertip contact point, F tip,jRepresents the contact force of the jth sensor in the 4*4 fingertip tactile array represents the derivative of the i-th joint angle to the j-th sensor in each fingertip tactile array, K tip Represents the mapping coefficient, which is used to gain the contact force mapping.

[0057] In the implementation manner of the present application, step S6 is specifically as follows: The impedance parameters are adjusted according to the stiffness of the grasped object to achieve variable impedance control, and it automatically switches to the admittance control mode when a sudden external force is detected (dF / dt > 100N / s).

[0058] The finger collaborative trajectory is corrected online based on the contact force distribution to ensure the stability of multi-finger enveloping grasping.

[0059] A safety boundary protection mechanism is set up to ensure the safe operation of the exoskeleton manipulator through joint angle-torque envelope constraints.

[0060] In the implementation manner of the present application, in step S6, the impedance parameter adjustment formula of the variable impedance control is:

[0061] Among them, K p is the stiffness parameter, K p0 is the set foundation stiffness parameter, is the stiffness variation coefficient, B d is the damping parameter, B d0 is the basic damping parameter set, Indicates F tip The second-order vector norm of ||x|| n is the standard form of the vector norm, that is, the magnitude of the vector in this vector space. The expression is:

[0062] Where n is the order and N is the number of elements in the vector, such as F tip Since it is a 4*4 fingertip tactile array, N=16.

[0063] The online correction formula of finger collaborative trajectory is:

[0064] in, is the corrected real-time reference angle of the i-th joint (where the superscript new represents real time), θ d,i is the original expected angle of the i-th joint, which comes from the intention trajectory generated by the electromyographic signal, neighbors is the set of collaborative joints, including the joints of all fingers involved in the enveloping grasp, θ d,jis the desired angle of the neighboring joint j associated with joint i.

[0065] H(zi) is used to convert the delayed contact state z i Converted into equivalent potential energy, H(z i )=0.5k p z i 2 , where k p is the virtual contact stiffness coefficient), K c is the synergistic gain, K h is the lag gain.

[0066] The joint angle-torque envelope constraint formula of the safety boundary protection mechanism is:

[0067] Among them, τ cmd,i is the final output control torque of the i-th joint, τ comp,i is the compensation torque of the i-th joint, τ i is the actual torque output at present, τ max is the maximum output torque limit, τ soft It is the buffer target torque when the joint approaches the safety limit, which reduces the joint impact by reducing the output torque, η=sigmoid(|θ i -θ max,i ∣) to achieve a smooth transition.

[0068] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any form, and any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for controlling an exoskeleton manipulator, characterized in that: The following steps are involved: S1: Construct a hand biomechanical model, define the degrees of freedom of the finger joints, and perform joint modeling based on the driving dynamics equations of the Bowden cable; S2: Deploy high-density sensors on the hand, including fingertip tactile arrays, joint micro-encoders, and surface electromyography sensors. The fingertip tactile array is used to measure contact force distribution, the joint micro-encoders are used to measure joint angles, and the surface electromyography sensors are used to collect electromyographic signals of the forearm flexor / extensor muscle groups. S3: Improve the hand-specific Bouc-Wen model, introduce inter-finger coupling terms to correct the hysteresis state equation, and identify the micro-hysteresis parameters through a hierarchical identification algorithm; S4: Design an adaptive collaborative control strategy to realize the fuzzy mapping from EMG signals to joint torques, and optimize the collaborative tracking error through the multi-joint parameter coupling adaptive law; S5: A modified dual Kalman filter is used to dynamically compensate for the mechanical resistance of the hand, including a state filter and a parameter filter, which are used to estimate the joint friction force and update the friction coefficient in real time, respectively; S6: Optimize closed-loop execution and safety strategies, including variable impedance control, finger collaborative trajectory optimization, and safety boundary protection mechanism.

2. The exoskeleton manipulator control method according to claim 1, characterized in that: In step S1, the finger joints have 14 degrees of freedom, and the dynamic equation of each joint is: ， Where i=1,2,…,14 represents the joint number, τ i is the driving torque of the i-th joint, J i is the inertia term, θ i is the joint angle, B i is the damping term, K i is the elastic term, H i (z i ) is the lag term, τ ext,i is the external torque term, F fric,i is the i-th joint friction term, which includes the coupling effect of fingertip contact friction and joint bearing friction.

3. The exoskeleton manipulator control method according to claim 2, characterized in that: In step S2, the fingertip tactile array is a 4*4 piezoresistive sensor, and the contact force F tip The distribution is calculated as: ， Among them, w j is the normalized weight, P j is the measurement value of the jth sensor, w j ∈[0,1],∑w j =1.

4. The exoskeleton manipulator control method according to claim 3, characterized in that: In step S3, the improved Bouc-Wen hysteresis state equation is: ， Among them, z i represents the hysteresis state variable of the i-th joint, C ij is the inter-finger coupling coefficient, which is calibrated through grasping experiments; The micro-hysteresis parameters are identified by a hierarchical identification algorithm, and the outer layer uses the PSO optimization algorithm to optimize α i , β i , γ i and n,α i is the hysteresis stiffness coefficient of the i-th joint, controlling the hysteresis linear part, β i is the nonlinear hysteresis damping parameter of the i-th joint, affecting the saturation rate, γ i is the nonlinear hysteresis shape parameter of the i-th joint, n∈[1,3], is the hysteresis nonlinear order; The inner layer uses the least squares method to update C in real time ij and F fric,i .

5. The exoskeleton manipulator control method according to claim 4, characterized in that: In step S4, the fuzzy mapping formula from electromyographic signal to joint torque is: ， Among them, a m is the myoelectric activation degree of the mth channel, a th is the individual activation threshold, K sEMG =0.5N*m / V, represents the electromyographic signal-torque conversion gain coefficient, which is determined by calibration experiments, θ max,i Represents the maximum allowed angle of the i-th joint, avoiding sudden changes in the mapping value when the joint angle approaches the limit.

6. The exoskeleton manipulator control method according to claim 4, characterized in that: In step S4, the collaborative tracking error is defined as: ， Where λ=0.3, w i is the weight coefficient, w i ∝1 / J i , N=14, θ i is the angle of the i-th joint, θ j is the angle of the jth joint; The improved adaptive algorithm is: ， in, represents the joint equivalent inertia adjusted in real time, where ^ represents the derivative, γJ is the adaptive gain, which is used to control the update rate of the inertia parameters, and e i is the tracking error of the i-th joint, J nom is the nominal inertia value, which is set according to the initial biomechanical model. max It is the maximum inertia value to prevent parameter overshoot.

7. The exoskeleton manipulator control method according to claim 4, characterized in that: In step S5, the prediction and update equations of the state filter are: ， Among them A i = diag[0.92, 0.85],A i is the state transfer matrix, which describes the dynamic attenuation characteristics of the friction state and is calibrated by step response experiments. represents the first derivative of the angle of the i-th joint, represents the first derivative of the angle of the i-th joint at the previous moment, is the friction state vector of the i-th joint, represents the friction state vector of the i-th joint at the previous moment, K f,i represents the Kalman gain matrix of the i-th joint state filter, F tip,i represents the contact force distribution of the i-th fingertip tactile array, , H i is the observation matrix, which maps the friction state to the sensor measurement space, where =0.01 rad / s, is the velocity normalization factor, B fric,i is the input coupling matrix, which is the excitation weight of the angular velocity on the friction state.

8. The exoskeleton manipulator control method according to claim 4, characterized in that: In step S5, the update equation of the parameter filter is: ， Among them, σ i is the friction coefficient, is the friction coefficient at the current moment, represents the friction coefficient at the previous moment, K σ,i is the parameter update gain, which controls the parameter convergence rate. is the model sensitivity matrix; The compensation torque formula for fingertip contact force feedforward compensation is: ， Among them, τ comp,i is the compensation torque of the ith joint, x j is the coordinate of the fingertip contact point, F tip,j represents the contact force of the jth sensor in the 4*4 fingertip tactile array, represents the derivative of the i-th joint angle to the j-th sensor in each fingertip tactile array, K tip Represents the mapping coefficient, which is used to gain the contact force mapping.

9. The exoskeleton manipulator control method according to claim 4, characterized in that: Step S6 is specifically as follows: The impedance parameters are adjusted according to the stiffness of the grasped object to achieve variable impedance control, and it automatically switches to the admittance control mode when a sudden external force is detected; The finger collaborative trajectory is corrected online based on the contact force distribution to ensure the stability of multi-finger envelope grasping; A safety boundary protection mechanism is set up to ensure the safe operation of the exoskeleton manipulator through joint angle-torque envelope constraints.

10. The exoskeleton manipulator control method according to claim 9, characterized in that: In step S6, the impedance parameter adjustment formula of the variable impedance control is: ， Among them, K p is the stiffness parameter, K p0 is the set foundation stiffness parameter, is the stiffness variation coefficient, B d is the damping parameter, B d0 is the basic damping parameter set, Indicates F tip The second-order vector norm of ; The online correction formula of finger collaborative trajectory is: ， in, is the corrected real-time reference angle of the i-th joint, θ d,i is the original expected angle of the i-th joint, which comes from the intention trajectory generated by the electromyographic signal, neighbors is the set of collaborative joints, including the joints of all fingers involved in the enveloping grasp, θ d,j is the expected angle of the neighboring joint j associated with joint i, H(z i ) is used to convert the delayed contact state z i Converted into equivalent potential energy, K c is the synergistic gain, K h is the lag term gain; The joint angle-torque envelope constraint formula of the safety boundary protection mechanism is: ， Among them, τ cmd,i is the final output control torque of the i-th joint, τ comp,i is the compensation torque of the i-th joint, τ i is the actual torque output at present, τ max is the maximum output torque limit, τ soft It is the buffer target torque when the joint approaches the safety limit, which reduces the joint impact by reducing the output torque, η=sigmoid(|θ i -θ max,i ∣) to achieve a smooth transition.

Images