CN118288294B - A robot visual servo and human-machine collaborative control method based on image variable admittance - Google Patents

Abstract

The invention discloses a robot vision servo and man-machine cooperative control method based on image admittance, which comprises the following steps: performing kinematic and dynamic modeling on the mechanical arm in the visual servo system to obtain a mechanical arm kinematic model and a dynamic model; calibrating a visual servo system through a chessboard method to obtain a conversion matrix between the mechanical arm end effector and the camera; constructing a first-order kinematic model and a second-order kinematic model of the visual servo; obtaining a dynamic model of the mechanical arm in a characteristic space; calculating and adjusting admittance parameters by using virtual positions of feature points in the image; determining a visual servo control law; the camera acquires image characteristic points of the two-dimensional code in real time, and the joint speed of the mechanical arm is obtained in real time according to a visual servo control law, so that the mechanical arm is subjected to motion control, and a visual servo process is completed. The invention solves the problem of inconsistent driving layers of the force sensor and the visual sensor, couples the visual sensor and the force sensor, and improves the flexibility of the system.

Description

A robot visual servo and human-machine collaborative control method based on image variable admittance

Technical Field

The invention relates to a robot visual servo and human-machine collaborative control method based on image variable admittance, and belongs to the technical field of robot visual servo.

Background Art

Visual servoing is a control method that uses visual features (perception) to control the robot for robot positioning or trajectory tracking. Compared with sensorless systems, visual servo systems have higher accuracy, flexibility and robustness when dealing with unstructured environments. The robot's force control can better adapt to environmental uncertainties and improve the accuracy and safety of robot contact operations.

At present, in the operation and maintenance of electrified railway contact network in my country, "detection" has been automated and partially intelligent, mainly using 6C system, inspection and patrol, component inspection and other methods and means to analyze and diagnose the technical status of the contact network. "Maintenance" is carried out according to the technical status, including temporary repair, comprehensive repair and precise measurement and repair, but the above three levels of maintenance basically rely on manual labor. Visual servo and human-machine collaboration are important technologies for realizing semi-automatic and full-automatic intelligent maintenance of contact network.

At present, most human-machine interactive control technologies are based on direct explicit closed-loop force control or indirect force control, such as impedance and admittance control. Multi-sensor fusion can realize multi-level and multi-space information processing, enhancing the flexibility and intelligence of the robot system. However, the combination of vision and force sensors is not an easy task. Some scholars have designed a control strategy based on the form of a task framework, which improves the shared control force by controlling the direction of the constraint, and the remaining direction of the movement is controlled by vision. Other scholars have proposed a hybrid vision/force control method to deal with the uncertainty of the camera and the constraint surface by tracking the image features of the end effector and applying orthogonal contact forces to the unknown surface. The above methods all use visual and force information for motion control, but there is a disadvantage that vision cannot correct the errors caused by the force control direction.

In response to the above problems, some scholars proposed an adaptive dynamic controller based on task space feedback to solve the contact problem under bottleneck constraints through vision/force hybrid control. The task space is divided into constraint space, image space and force space to complete the visual servo operation and force trajectory tracking control tasks. Some scholars have also designed first-order and second-order impedance control laws based on image vision. In the above control law, the elastic torque component of the mixed impedance equation is defined on the image plane for the design and calculation of visual errors. Although the above method divides the task space into image feature space and designs a controller based on visual errors, its compliance is still implemented in Cartesian space. Unlike vision/force hybrid control in Cartesian space or joint space, scholars proposed a generalized framework for vision/impedance hybrid control in feature space. This framework is defined in the task space of the visual servo system without considering the selection of visual features. Compliance is achieved by selecting a constant admittance parameter in the feature space.

Summary of the invention

In order to overcome the shortcomings of hybrid vision/force control in the prior art that vision cannot correct the errors caused by the force control direction, and the problems of poor system flexibility and safety in constant admittance control, the present invention aims to provide a robot visual servo and human-machine collaborative control method based on image variable admittance.

The present invention provides a technical solution to solve the above technical problems: a robot visual servo and human-machine collaborative control method based on image variable admittance, comprising the following steps:

Step S100, performing kinematic and dynamic modeling on the robot arm in the visual servo system to obtain a kinematic model and a dynamic model of the robot arm, and setting a coordinate system for a reference frame of the visual servo system;

Step S200, calibrating the visual servo system by a chessboard method to obtain a transformation matrix between the end effector of the robot arm and the camera;

Step S300, establishing the kinematic and dynamic models of camera vision, and combining the kinematic model and dynamic model of the robot arm to construct the first-order kinematic model and the second-order kinematic model of the visual servo;

Step S400, establishing a dynamic model of the robot arm in the image feature space, combining the robot arm kinematics, dynamic model and the first-order kinematic model and second-order kinematic model of the visual servo to obtain the dynamic model of the robot arm in the feature space;

Step S500, combining the logistic sigmoid function, using the virtual position of the feature point in the image to calculate and adjust the admittance parameter in real time;

Step S600, determining a visual servoing control law according to an error between a desired feature point and a current feature point;

Step S700: The camera collects the image feature points of the QR code in real time, and obtains the joint speed of the robot arm in real time according to the visual servo control law, thereby controlling the motion of the robot arm and completing the visual servo process.

A further technical solution is that the kinematic model of the robotic arm is:

Where: , , They are the joint position vector, velocity vector and acceleration vector in the joint space respectively; is the Jacobian matrix of the robot arm; , , are the position vector, velocity vector and acceleration vector in the operation space respectively, is the coordinate system mapping function.

A further technical solution is that the mechanical arm dynamics model is:

Where: , , They are the joint position vector, velocity vector and acceleration vector in the joint space respectively; is the inertia matrix of the manipulator; are the Coriolis and centripetal matrices; is the gravity vector; are the Coulomb, viscosity and static friction vectors; is the external torque; For control input.

A further technical solution is that the specific process of step S200 includes:

Step S210: using a chessboard calibration plate with N points and randomly placing it in the field of view of the RealSense camera;

Step S220: Use the RealSense camera to collect images of the chessboard calibration plate;

Step S230, calculating the pixel coordinates of the centers of N dots in the image;

Step S240, recording the three-dimensional position and posture of the end effector of the robot arm at this time in the form of a translation vector and a rotation vector;

Step S250, repeating steps S220 to S240 a total of 16 times;

Step S260: Solve the 2D-3D data to find the coordinate transformation relationship between the chessboard calibration plate and the camera ;

Step S270: According to the coordinate conversion relationship , calculate the transformation matrix from the end effector of the robot to the camera .

A further technical solution is that the first-order kinematic model of the visual servo is:

Where: is the image feature point velocity, is the image interaction matrix, is the camera's movement speed.

A further technical solution is that the second-order kinematic model of the visual servo is:

Where: is the transformation matrix from the robot end effector to the camera; is the Jacobian matrix of the robot arm; is the image feature Jacobian matrix; , are the velocity vector and acceleration vector in the joint space respectively; is the image interaction matrix; is the acceleration vector of the image feature point; is the first-order derivative of the image interaction matrix; is the first-order derivative of the transformation matrix from the robot end effector to the camera; is the first-order derivative of the Jacobian matrix of the robot.

A further technical solution is that the dynamic model of the robot arm in the feature space is:

in:

Where: is the inverse of the projection of the manipulator inertia matrix into the camera frame; is the characteristic Jacobian matrix; is the Jacobian matrix of the robot arm; is the transformation matrix from the robot end effector to the camera; is the acceleration vector of the image feature point; is the moment in the image feature space; is the external contact force in the image feature space; is the external disturbance in the image feature space; is the transpose of the Jacobian matrix of the robot; is the transformation matrix from the robot end effector to the camera; is the force in the camera coordinate system.

A further technical solution is that the specific process of step S500 is: select the logistic sigmoid function in combination with the admittance formula in the feature space and the feature point distance difference, and adjust the damping through the following s-shaped function and stiffness Variable admittance parameter;

Where: and It is a parameter that affects the growth rate and inflection point of the curve; is the logistic sigmoid function; is the maximum value of the admittance parameter; is the minimum value of the admittance parameter; is the total length of the virtual path; It is the length of the virtual path between the current feature point and the expected feature point.

A further technical solution is that the damping is calculated and adjusted in step S500 When: Select is the maximum value of the damping parameter, is the minimum value of the damping parameter, , , the damping coefficient is adjusted by the sigmoid function of this set of parameters;

Calculating and adjusting stiffness When: Select is the maximum value of the stiffness parameter, is the minimum value of the stiffness parameter, , , the stiffness coefficient is adjusted by a sigmoid function of this set of parameters.

A further technical solution is that the specific process of step S600 includes:

Step S610, establishing an admittance control ratio between the external contact force and the characteristic point error;

Where: , , are virtual mass, damping and stiffness respectively; is the feature point error; is the acceleration of the feature point error; is the speed of feature point error; is the expected feature point set initially, is the feature point moved by the external contact force; is the external contact force in the image feature space;

Step S620: Select As the error between the current feature point and the expected feature point, is the image feature point;

Step S630: Acceleration of feature point error Performing the first and second integrations can obtain the speed of the characteristic point error and feature points moved by external contact forces ;

Step S640: Determine the visual servo control law as:

Where: is the camera's movement speed; is the speed of feature point error; is the controller gain; is the pseudo-inverse of the image interaction matrix; is the error between the current feature point and the expected feature point.

The present invention has the following beneficial effects:

1. Different from the visual/force hybrid control in Cartesian space or joint space, this invention solves the problem of inconsistent driving layers of force sensors and visual sensors;

2. Different from variable admittance control in Cartesian space or joint space, this invention couples visual sensors and force sensors to improve the flexibility of the system;

3. Compared with the invariant admittance control of vision/force hybrid in feature space, the introduction of variable admittance can adjust the admittance parameters in real time, improving the dynamic characteristics, human-computer interaction and safety of the system.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a flow chart of the present invention;

FIG2 is a curve showing a change in admittance parameters in a variable admittance system according to an embodiment of the present invention;

Figure 2(a) shows the change of damping parameters, and Figure 2(b) shows the change of stiffness parameters;

3 is a schematic diagram of the execution effect of the visual servo stage of the control method for different admittance parameters in one embodiment of the present invention;

FIG3(a) shows the implementation effect of low damping, high stiffness and constant admittance, FIG3(b) shows the implementation effect of high damping, low stiffness and constant admittance, FIG3(c) shows the implementation effect of high damping, high stiffness and constant admittance, and FIG3(d) shows the implementation effect of variable admittance.

FIG4 is a schematic diagram of the execution results of a control method for four different admittance parameters in one embodiment of the present invention;

Among them, Figure 4(a) shows the execution result of visual servoing completion time and its error bar, Figure 4(b) shows the execution result of speed and angular velocity and their error bars during the implementation process, Figure 4(c) shows the execution effect of total task completion time and its error bar, and Figure 4(d) shows the execution effect of force and its error bar during the task.

DETAILED DESCRIPTION

The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

As shown in FIG1 , a robot visual servo and human-machine collaborative control method based on image variable admittance specifically includes the following steps:

Step S100, performing kinematic and dynamic modeling on the robot arm in the visual servo system to obtain a kinematic model and a dynamic model of the robot arm, and setting a coordinate system for a reference frame of the visual servo system;

The kinematic model of the robot arm is:

Where: , , They are the joint position vector, velocity vector and acceleration vector in the joint space respectively; is the Jacobian matrix of the robot arm; , , are the position vector, velocity vector and acceleration vector in the operation space respectively, is the coordinate system mapping function.

The mechanical arm dynamics model is:

in:

Where: , , They are the joint position vector, velocity vector and acceleration vector in the joint space respectively; is the inertia matrix of the manipulator; are the Coriolis and centripetal matrices; is the gravity vector; are the Coulomb, viscosity and static friction vectors; is the external torque; is the external torque in the end-effector coordinate system; is the control input;

Step S200, calibrating the visual servo system by a chessboard method to obtain a transformation matrix between the end effector of the robot arm and the camera;

Step S210: using a chessboard calibration plate with N points and randomly placing it in the field of view of the RealSense camera;

Step S220: Use the RealSense camera to collect images of the chessboard calibration plate;

Step S230, calculating the center pixel coordinates of N dots in the image;

Step S240, recording the three-dimensional position and posture of the end effector of the robot arm at this time in the form of a translation vector and a rotation vector;

Step S250, repeating steps S220 to S240 a total of 16 times;

Step S260: Solve the 2D-3D data to find the coordinate transformation relationship between the chessboard calibration plate and the camera ;

Step S270: According to the coordinate conversion relationship , calculate the transformation matrix from the end effector of the robot to the camera ;

Step S300, establishing the kinematic and dynamic models of camera vision, and constructing the first-order kinematic model and the second-order kinematic model of visual servoing in combination with the kinematic model and the dynamic model of the robot arm;

The first-order kinematic model of visual servoing is:

Where: is the image feature point velocity, is the image interaction matrix, is the camera's movement speed;

The second-order kinematic model of visual servoing is:

Where: is the transformation matrix from the robot end effector to the camera; is the Jacobian matrix of the robot arm; is the image feature Jacobian matrix; , are the velocity vector and acceleration vector in the joint space respectively; is the image interaction matrix; is the acceleration vector of the image feature point; is the first-order derivative of the image interaction matrix; is the first-order derivative of the transformation matrix from the robot end effector to the camera; is the first-order derivative of the Jacobian matrix of the robot;

Step S400: Establish a dynamic model of the robot in the image feature space, and obtain the dynamic model of the robot in the feature space by combining the robot kinematics, dynamic model, and the first-order kinematic model and second-order kinematic model of the visual servo, which is specifically expressed as:

in , and are the torque transformation matrix from the camera coordinate system to the end effector coordinate system and the torque acting in the camera coordinate system. Combined with the characteristic Jacobian matrix, the torsion transformation matrix is related to the wrench transformation matrix as follows: ;

Rewrite it to get the dynamic model of the robot arm in the feature space;

in:

Where: is the inverse of the projection of the manipulator inertia matrix into the camera frame; is the characteristic Jacobian matrix; is the Jacobian matrix of the robot arm; is the transformation matrix from the robot end effector to the camera; is the acceleration vector of the image feature point; is the moment in the image feature space; is the external contact force in the image feature space; is the external disturbance in the image feature space; is the transpose of the Jacobian matrix of the robot; is the transformation matrix from the robot end effector to the camera; is the force in the camera coordinate system;

Step S500, combining the logistic sigmoid function, using the virtual position of the feature point in the image to calculate and adjust the admittance parameter in real time;

Select the logistic sigmoid function, which is specifically expressed as:

Where: and It is a parameter that affects the growth rate and inflection point of the curve;

Combined with the admittance formula in the feature space and the distance difference of the feature points, the damping is adjusted by the following s-shaped function and stiffness Variable admittance parameter;

Where: and It is a parameter that affects the growth rate and inflection point of the curve; is the logistic sigmoid function; is the maximum value of the admittance parameter; is the minimum value of the admittance parameter; is the total length of the virtual path; is the length of the virtual path between the current feature point and the expected feature point;

Calculating and adjusting damping When: Select is the maximum value of the damping parameter, is the minimum value of the damping parameter, , , the damping coefficient is adjusted by the sigmoid function of this set of parameters, and the result is shown in Figure 2(a);

Calculating and adjusting stiffness When: Select is the maximum value of the stiffness parameter, is the minimum value of the stiffness parameter, , , the stiffness coefficient is adjusted by the sigmoid function of this set of parameters, and the result is shown in Figure 2(b);

Step S600, determining a visual servoing control law according to an error between a desired feature point and a current feature point;

Step S610, establishing an admittance control ratio between the external contact force and the characteristic point error;

Where: , , are virtual mass, damping and stiffness respectively; is the feature point error; is the acceleration of the feature point error; is the speed of feature point error; is the expected feature point set initially, is the feature point moved by the external contact force; is the external contact force in the image feature space;

Step S620: Select As the error between the current feature point and the expected feature point, is the image feature point;

is the expected feature point set initially, is the feature point moved by the external contact force;

Step S630: Acceleration of feature point error Performing the first and second integrations can obtain the speed of the characteristic point error and feature points moved by external contact forces ;

Step S640: Determine the visual servo control law as:

Where: is the camera's movement speed; is the speed of feature point error; is the controller gain; is the pseudo-inverse of the image interaction matrix; is the error between the current feature point and the expected feature point;

Step S700: The camera collects the image feature points of the two-dimensional code in real time, and obtains the joint speed of the robot arm in real time according to the visual servo control law, thereby controlling the motion of the robot arm and completing the visual servo process;

Specifically, programming is performed on the PC according to the designed visual servo controller, and the manipulator, torque sensor, and Visp platform communicate with each other through ROS. The embodiment is based on the background of overhead line bolt maintenance, and aims to improve the efficiency and safety of the overhead line maintenance process.

The embodiment includes three stages. In the first stage, the maintenance target is identified and visual servoing is performed to the maintenance position. In the second stage, the maintenance robot arm is moved to the next maintenance point through human-machine collaboration. In the third stage, the next maintenance target is identified and visual servoing is performed to the maintenance position.

Figure 3 shows the experimental results. Figure 3(a) is an experiment with low damping and high stiffness. Its visual servo converges faster, but there is overshoot. Figure 3(b) is an experiment with high damping and low stiffness. Its visual servo converges the slowest. Figure 3(c) is an experiment with high damping and high stiffness. Due to the existence of high stiffness, its visual servo convergence speed is faster than that of high damping and low stiffness, but slower than that of high stiffness. Figure 3(d) is an experiment with variable admittance. Its visual servo converges the fastest and there is no overshoot.

Figure 4 is a display of the experimental results. Figure 4 (a) represents the convergence time of visual servoing, and the variable admittance control has the best effect; Figure 4 (b) represents the speed during the operation of visual servoing, and the variable admittance control speed is only lower than the low damping and high impedance control; Figure 4 (c) represents the total task time, and the variable admittance control has the shortest task completion time; Figure 4 (d) represents the interactive force during the human-computer interaction process, and the interactive force of the variable admittance control is only larger than that of the high damping and low stiffness; Based on the above experimental performance indicators, the variable admittance has the best dynamic response in the pure visual servoing process, and has good human-computer interactivity in the entire experimental task. This result verifies the effectiveness of the present invention.

Compared with the existing technology, the advantages of the present invention are: (1) Different from the vision/force hybrid control in Cartesian space or joint space, the present invention solves the problem of inconsistent driving layers of force sensors and vision sensors; (2) Different from the variable admittance control in Cartesian space or joint space, the present invention couples the vision sensor and the force sensor, thereby improving the flexibility and human-computer interactivity of the system; (3) Compared with the invariant admittance control of vision/force hybrid in feature space, the introduction of variable admittance can adjust the admittance parameters in real time, thereby improving the dynamic characteristics, human-computer interactivity and safety of the system.

The above description is not intended to limit the present invention in any form. Although the present invention has been disclosed through the above embodiments, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are within the scope of the technical solution of the present invention.

Claims (7)

1. A robot visual servo and human-machine collaborative control method based on image variable admittance, characterized in that it includes the following steps: Step S100, performing kinematic and dynamic modeling on the robot arm in the visual servo system to obtain a kinematic model and a dynamic model of the robot arm, and setting a coordinate system for a reference frame of the visual servo system; Step S200, calibrating the visual servo system by a chessboard method to obtain a transformation matrix between the end effector of the robot arm and the camera; Step S210: using a chessboard calibration plate with N points and randomly placing it in the field of view of the RealSense camera; Step S220: Use the RealSense camera to collect images of the chessboard calibration plate; Step S230, calculating the pixel coordinates of the centers of N dots in the image; Step S240, recording the three-dimensional position and posture of the end effector of the robot arm at this time in the form of a translation vector and a rotation vector; Step S250, repeating steps S220 to S240 a total of 16 times; Step S260, solving the 2D-3D data to obtain the coordinate transformation relationship ^c T _o between the chessboard calibration plate and the camera; Step S270, calculating the transformation matrix ^c _Te from the robot end effector to the camera according to the coordinate transformation relationship ^c T _o ; Step S300, establishing the kinematic and dynamic models of camera vision, and combining the kinematic model and dynamic model of the robot arm to construct the first-order kinematic model and the second-order kinematic model of the visual servo; Step S400, establishing a dynamic model of the robot arm in the image feature space, and combining the robot arm kinematics, dynamic model and the first-order kinematic model and second-order kinematic model of the visual servo to obtain the dynamic model of the robot arm in the image feature space; in: Where: _Mc (q) ^-1 is the inverse of the projection of the manipulator inertia matrix in the camera frame; _Js is the image feature Jacobian matrix; ^eJe is _the manipulator Jacobian matrix; ^cTe is the transformation _matrix from the manipulator end effector to the camera; is the acceleration vector of the image feature point; _fs is the torque in the image feature space; _fsext is the external contact force in the image feature space; _fq is the external disturbance in the image feature space; is the transpose of the Jacobian matrix of the robot; is the transformation matrix from the end effector of the robot to the camera; ^c f _c is the force in the camera coordinate system; are the Coriolis matrix and the centripetal matrix; g(q) is the gravity vector; are the Coulomb, viscosity and static friction vectors; τ is the control input; q, They are the joint position vector and velocity vector in the joint space respectively; Step S500, combining the logistic sigmoid function, using the virtual position of the feature point in the image to calculate and adjust the admittance parameter in real time; The logistic sigmoid function is selected to combine the admittance formula in the feature space and the distance difference of the feature points to adjust the variable admittance parameters of damping _Ds and stiffness _Ks ; Where: γ ₁ and γ ₂ are parameters that affect the growth rate and inflection point of the curve; σ( _es ) is the logistic sigmoid function; V _max is the maximum value of the admittance parameter; V _min is the minimum value of the admittance parameter; _ds is the total length of the virtual path; _es is the length of the virtual path between the current feature point and the expected feature point; Step S600, determining a visual servoing control law according to an error between a desired feature point and a current feature point; Step S700: The camera collects the image feature points of the QR code in real time, and obtains the joint speed of the robot arm in real time according to the visual servo control law, thereby controlling the motion of the robot arm and completing the visual servo process. 2. According to the method of robot visual servo and human-machine collaborative control based on image variable admittance in claim 1, it is characterized in that the kinematic model of the robot arm is: x＝P(q) Where: q, are the joint position vector, velocity vector and acceleration vector in the joint space respectively; ^e J _e is the Jacobian matrix of the robot; x, are the position vector, velocity vector and acceleration vector in the operation space respectively, and P(q) is the coordinate system mapping function; is the first-order derivative of the Jacobian matrix of the robot. 3. According to the method of robot visual servo and human-machine collaborative control based on image variable admittance in claim 1, it is characterized in that the mechanical arm dynamics model is: Where: q, are the joint position vector, velocity vector and acceleration vector in the joint space respectively; M(q) is the inertia matrix of the manipulator; are the Coriolis matrix and the centripetal matrix; g(q) is the gravity vector; are the Coulomb, viscosity and static friction vectors; τ _e is the external torque; τ is the control input. 4. According to the method of robot visual servo and human-machine collaborative control based on image variable admittance in claim 1, it is characterized in that the first-order kinematic model of the visual servo is: Where: is the speed of the image feature points, _Ls is the image interaction matrix, and _vc is the movement speed of the camera. 5. The robot visual servo and human-machine collaborative control method based on image variable admittance according to claim 4, characterized in that the second-order kinematic model of the visual servo is: Where: ^c T _e is the transformation matrix from the end effector of the robot to the camera; ^e J _e is the Jacobian matrix of the robot; J _s is the Jacobian matrix of the image feature; are the velocity vector and acceleration vector in the joint space respectively; _Ls is the image interaction matrix; is the acceleration vector of the image feature point; is the first-order derivative of the image interaction matrix; is the first-order derivative of the transformation matrix from the robot end effector to the camera; is the first-order derivative of the Jacobian matrix of the robot arm; _fq is the external disturbance in the image feature space. 6. A robot visual servo and human-machine collaborative control method based on image variable admittance according to claim 1, characterized in that, when calculating and adjusting the damping _Ds in step S500: V _max = 40 is selected as the maximum value of the damping parameter, V _min = 10 is selected as the minimum value of the damping parameter, γ ₁ = 25, γ ₂ = 2.5, and the damping coefficient is adjusted by the logistic sigmoid function of this group of parameters; When calculating and adjusting the stiffness _Ks : select V _max = 200 as the maximum value of the stiffness parameter, V _min = 100 as the minimum value of the stiffness parameter, γ ₁ = 50, γ ₂ = 1.5, and the stiffness coefficient is adjusted by the logistic sigmoid function of this set of parameters. 7. The robot visual servo and human-machine collaborative control method based on image variable admittance according to claim 1, characterized in that the specific process of step S600 includes: Step S610, establishing an admittance control ratio between the external contact force and the characteristic point error; e＝ ^sd - ^sw Where: _Hs , _Ds , _Ks are virtual mass, damping and stiffness respectively; e is the characteristic point error; is the acceleration of the feature point error; is the speed of feature point error; ^sd is the expected feature point set initially, ^sw is the feature point moved by the external contact force; _fsext is the external contact force in the image feature space; Step S620, selecting e _s =ss ^w as the error between the current feature point and the expected feature point, where s is the image feature point; Step S630: Acceleration of feature point error Performing the first and second integrations can obtain the speed of the characteristic point error and the characteristic point s ^w moved by the external contact force; Step S640: Determine the visual servo control law as: Where: v _c is the camera's movement speed; is the speed of the characteristic point error; α is the controller gain; is the pseudo-inverse of the image interaction matrix; _es is the error between the current feature point and the expected feature point.