WO2020118730A1

WO2020118730A1 - Compliance control method and apparatus for robot, device, and storage medium

Info

Publication number: WO2020118730A1
Application number: PCT/CN2018/121338
Authority: WO
Inventors: 欧勇盛; 段江哗; 徐升; 王志扬; 金少堃; 田超然; 王煜睿; 熊荣; 江国来; 吴新宇
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2018-12-14
Filing date: 2018-12-14
Publication date: 2020-06-18

Abstract

A compliance control method for a robot. The method comprises: acquiring demonstration data of a demonstration motion; calculating a motion equation of the demonstration motion according to motion data in the demonstration data, and calculating variable impedance parameters of the demonstration motion according to interaction force data in the demonstration data at the same time; and controlling the operation according to the motion equation and variable impedance parameters, so that the manual programming during robot compliance control is omitted, the difficulty for using robots is lowered, and the compliance and accuracy of the robot control are improved, thereby improving the generalization ability, intelligence, and control effect of robots. Also related is a compliance control apparatus for a robot, a device, and a storage medium.

Description

Robot compliance control method, device, equipment and storage medium

Technical field

The invention belongs to the field of computer technology, and particularly relates to a robot compliance control method, device, equipment and storage medium.

Background technique

In the application of robots at this stage, especially in industrial applications, the trajectory of the robotic arm is generally pre-defined by the user, or a certain task environment is preset, and then the robot or the robotic arm can be repeatedly executed according to the plan. The robotic arm operating in this mode cannot face environmental changes or sudden disturbances. For the realization of complex scenarios or more difficult tasks, this mode also requires more arduous manual programming. For ordinary workers, the use threshold is high (for example: to be able to program robots). More importantly, this robot control mode does not imply human operation habits, nor is it as flexible as human hands. In order to effectively lower the threshold for the use of robots and better achieve human-machine collaborative interaction, the robotic arm or robot should have learning capabilities and be more flexible and compliant. The robot "Imitation Learning" (Imitation Learning) or "Teaching Learning" (Programming by Demonstration) is an important method to solve this problem.

Generally, the compliant behavior of a robot includes two aspects of action and force, so the learning of compliant behavior also includes two aspects of action learning and force learning.

In the field of robot compliance control, previous research work has focused on the artificial design of controllers (such as force-position hybrid control, impedance control, collision detection feedback controller, etc.) and passive compliance mechanism design. The above-mentioned design method of the compliant controller has a complicated parameter adjustment process, and does not have the generalization ability to adapt to the new situation. The study of robots to obtain compliance control strategies by learning human compliance behavior can simplify the complex parameter adjustment process and lower the threshold for robot use (workers only need to provide correct human teaching to allow robots to have corresponding compliance behavior, without Users need to have the relevant technical foundation of programming and robot control).

The study of robots' compliance control strategies obtained by learning human compliance behavior belongs to the frontier field. Most of the teaching learning control will independently study motion trajectory learning and force learning. For example, Seyed Mohammad Khansari-Zadeh proposed a method to learn the trajectory of movement (the article "Learning Stable Nonlinear Dynamics Systems With Gaussian Mixture Models" published in 2011 IEEE Transactions on Robotics). When this method was first proposed, the dynamic system was modeled by Gaussian Mixture Models, and constraints based on Lyapunov stability were also derived to ensure that the motion converged to the target. Other imitation learning methods of trajectory movements have emerged in the development in the following years, but the use of dynamic system modeling and the use of Lyapunov stability to constrain these two characteristics are basically the common characteristics of various methods. Calinon proposed a learning method for deriving different interaction forces based on the covariance of the perturbation of the teaching position, but this method is weird in teaching and is not conducive to learning together with the trajectory.

From the existing data, the motion trajectory and force are regarded as two components of compliant behavior, and there are very few mature schemes that use the two for robot learning and control of compliant behavior. An article published in 2017 by Autonomous Robots, "Learning potentials from human demonstrations with encapsulated dynamics and compliant behaviors," proposes a joint variable impedance control strategy based on potential functions and dissipative fields. This method requires artificially designing multiple groups through prior knowledge. Based on the parameters of the task, this method is strongly constructed, and can only be trained offline, which is inefficient.

Some of the patents that have been applied for or granted are also related to the mentioned fields. In the patent document titled "A Robotic Imitation Based on Gaussian Process", a robotic imitation learning method based on Gaussian process is disclosed. The Gaussian process is also a regression algorithm, similar to the Gaussian mixture model. This scheme uses the Gaussian process to model and learn the robot motion. In the patent document entitled "A Robot Chinese Character Writing Learning Method Based on Trajectory Imitation", a method of imitation learning based on trajectory matching is introduced into the learning of robot writing skills, and the strokes of Chinese characters are divided and passed A method of coding learning and reconstruction of teaching data by multiple Gaussian mixture models. In the patent document entitled "Hand-teaching robotic arm system and method with imitation learning mechanism", a robotic arm system with imitation learning function is disclosed, and imitation learning based on feedforward neural network is given.模方法。 Modal method. In the patent document titled "A device and method for robotic force-control teaching imitation learning", it is disclosed that force feedback information is introduced into the teaching data, and the hidden Markov model is used to perform the teaching data. Modeling coding method.

In summary, the existing robot compliance control method independently models and learns the motion trajectory and force, and the learning effect is not good, which leads to inaccurate control results; based on Gaussian mixture model, Gaussian process and other offline regression methods to For imitation learning, the training time required is relatively long, and the training efficiency is low; the stability of the control cannot be guaranteed, and there may be situations where the robot interaction force is too large and hurts people.

Summary of the invention

The object of the present invention is to provide a robot compliance control method, device, equipment and storage medium, aiming to solve the problems of inaccurate control results and poor control effects caused by poor compliance of the existing robot compliance control methods.

In one aspect, the present invention provides a robot compliance control method, which includes the following steps:

Acquiring teaching data of a teaching movement, wherein the teaching data includes at least the movement data and interaction force data of the teaching movement;

Calculating the motion equation of the teaching motion based on the motion data in the teaching data, and simultaneously calculating the variable impedance parameter of the teaching motion based on the interaction force data in the teaching data, wherein the variable impedance The parameters include at least variable stiffness parameters and variable damping parameters;

The operation is controlled according to the equation of motion and the variable impedance parameter.

In another aspect, the present invention provides a robot compliance control device, the device including:

A data acquisition unit for acquiring teaching data of the teaching movement, wherein the teaching data includes at least the movement data and the interaction force data of the teaching movement;

A parameter calculation unit, configured to calculate the motion equation of the teaching motion based on the motion data in the teaching data, and at the same time calculate the variable impedance parameter of the teaching motion based on the interaction force data in the teaching data, Wherein, the variable impedance parameter includes at least a variable stiffness parameter and a variable damping parameter; and

The operation control unit is used for controlling operation according to the motion equation and the variable impedance parameter.

On the other hand, the present invention also provides a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, which is implemented when the processor executes the computer program The steps of the robot compliance control method as described.

On the other hand, the present invention also provides a computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, implements the steps of the robot compliance control method.

The invention obtains the teaching data of the teaching movement, calculates the motion equation of the teaching movement based on the movement data in the teaching data, and simultaneously calculates the teaching according to the interaction force data in the teaching data The variable impedance parameter of the movement controls the operation according to the motion equation and the variable impedance parameter, thereby reducing manual programming during the robot compliance control process, lowering the threshold for robot use, and improving the flexibility and accuracy of robot control , And further improve the robot's generalization ability, intelligence and control effect.

BRIEF DESCRIPTION

1 is an implementation flowchart of a robot compliance control method provided in Embodiment 1 of the present invention;

2 is a schematic structural diagram of a robot compliance control device according to Embodiment 2 of the present invention;

3 is a schematic structural diagram of a robot compliance control device according to Embodiment 3 of the present invention; and

4 is a schematic structural diagram of a computing device according to Embodiment 4 of the present invention.

[Correction based on Rule 91 01.01.2019]
5 is a schematic structural diagram of a robot compliance control device according to Embodiment 2 of the present invention; and

[Correction based on Rule 91 01.01.2019]
6 is a schematic structural diagram of a computing device according to Embodiment 3 of the present invention.

detailed description

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not intended to limit the present invention.

The following describes the specific implementation of the present invention in detail with reference to specific embodiments:

Example one:

FIG. 1 shows the implementation flow of the robot compliance control method provided in Embodiment 1 of the present invention. For convenience of description, only the parts related to the embodiment of the present invention are shown. The details are as follows:

In step S101, the teaching data of the teaching movement is acquired.

The embodiments of the present invention are suitable for automatic control of robots. Robots include a series of robot products that are not limited to robotic arms, humanoid robots, etc. with joints, links, and other structures, and can achieve telescopic and grasping actions. Among them, the teaching data may include at least motion data and interaction force data of the teaching movement. Therefore, learning of the teaching movement may include action learning and force learning (ie, variable stiffness parameter and variable damping parameter learning).

In the embodiment of the present invention, the motion data may include position data and speed data of a preset point of the robot (eg, end effector), or the motion data may include angle and angle of a preset angle of the robot (eg, joint angle) Acceleration. In addition, the motion data may also include one or more other parameters that can be used to completely describe the teaching motion, which is not limited by the present invention.

As an example, FIG. 2 shows a diagram of teaching a robot. As shown in FIG. 2, during teaching, the teacher grasps the end effector of the robot with one hand and moves in a plane or space. Make a trajectory, and the other hand exerts a teaching force at the end. The robot collects teaching data through its own motion capture system and a six-dimensional force sensor mounted on the wrist.

For example, when the motion data includes position data and velocity data of a preset sampling point of the robot (eg, end effector or end, etc.), the position data and interaction force data of the end are sampled at time intervals to obtain a series of samples Point data

Where i=1,...,N _traj ,N _traj represents the number of teaching trajectories, k=1,...,N ⁱ ,N ⁱ represents the number of sampling points in the teaching (taken every other time interval Once),

That is, the end position of the k-th sampling point of the i-th trajectory, and the teaching force of the k-th sampling point of F _k .

As another example, during teaching, the teacher controls the robot through the remote control or the teach pendant to perform the teaching operation, or teaches by hand. The robot records the teaching data according to the teaching operation.

As yet another example, during teaching, the instructor personally completes the teaching movement task. The teaching data is collected by the robot's motion catcher, data glove, and force sensor according to the teaching movement.

Preferably, if the motion data includes position data and speed data of a preset point of the robot (for example, an end effector), when acquiring the teaching data of the teaching motion, the position data related to the teaching motion may be obtained first, Interaction force data and time data, and then calculate the speed data related to the teaching movement based on the position data and the time data, thereby obtaining the movement data of the teaching movement.

Preferably, if the motion data includes the angle and angular acceleration of the preset angle of the robot (for example, the joint angle), when acquiring the teaching data of the teaching motion, the angle data and interaction force related to the teaching motion may be obtained first Data and time data, and then calculate the angular acceleration data related to the teaching movement based on the angle data and the time data, thereby obtaining the movement data of the teaching movement.

In step S102, the motion equation of the teaching motion is calculated based on the motion data in the teaching data, and at the same time, the variable impedance parameter of the teaching motion is calculated based on the interaction force data in the teaching data.

In the embodiment of the present invention, the variable impedance parameter may include at least a variable stiffness parameter and a variable damping parameter. At the same time, the trajectory and force are learned, so as to improve the learning effect and thus the accuracy of the control results.

Preferably, when calculating the motion equation of the teaching motion based on the motion data in the teaching data, the preset neural network model can be trained using the motion data to obtain the motion equation of the teaching motion, and the neural network can be trained according to the motion equation The model is updated online, thereby improving the calculation efficiency of the motion equation, facilitating the subsequent use of the motion equation, and adapting to the needs of real-time online learning, thereby improving the learning effect.

Among them, preferably, when using the motion data to train the preset neural network model, the motion data can be incrementally learned one by one or block by block to obtain the motion equation of the teaching motion, thereby improving the accuracy of the motion equation Sex, thereby improving learning effectiveness.

The neural network model can be a support vector machine (Support Vector Machine, SVM), online sequence over-limit learning machine and other models that can be incrementally online learning, or other incremental online learning models, such as incremental support vector machine ( ISVM), etc., the present invention does not limit this. Among them, because compared with other online learning algorithms, the online sequence over-limit learning machine has the characteristics of fast learning speed, strong generalization ability, and simple implementation. Therefore, preferably, the neural network model is an online sequence over-limit learning machine, that is, use The motion data trains the online sequence overrun learning machine, thereby improving the training efficiency.

Among them, when training the online sequence over-limit learning machine, in terms of motion, the input and output are the position and speed (or the angle and angular acceleration of the joint angle) of the sampling point (for example, the robot end effector), so the online The input and output of the sequence overrun learning machine should have the same dimension, that is, the same number of neurons d. If you consider the movement in the two-dimensional plane, d = 2, if you consider the movement in the three-dimensional space, d = 3.

As an example, FIG. 3 shows an exemplary structure of an online sequence overrun learning machine. As shown in FIG. 3, assuming that the activation function of the hidden layer of the online sequence overrun learning machine is g, then the online sequence overrun learning we want to learn The machine (ie, the model to be learned) can be expressed as

Among them, the number of hidden layer neurons is

For the offset of the hidden layer,

Is the weight of the hidden layer, the dimension is

Is the weight of the output layer, the dimension is

Among them, in the training process of the online sequence over-limit learning machine, W and b are randomly generated and fixed. The training process only needs to determine the weight of the output layer.

Optimization process to achieve.

among them,

Indicates the target output in the teaching data. Since w and b are randomly generated and fixed, H is also fixed. The goal of training is to find the optimal set of output layer weights

Make

Get the minimum value.

Among them, although the activation function g generally selects the sigmoid function (sigmoid function) or the hyperbolic tangent function (tanh function), the modified sigmoid function can also be used, for example,

However, as long as it is satisfied

And the monotonically increasing continuous and continuously differentiable functions all meet the requirements of the activation function, and are not limited here.

The training goal of the online sequence overrun learning machine is to find a set of optimal output layer weights

Use the least squares method to get

Among them, H ⁺ is the Moore-Penrose generalized inverse matrix of the matrix H. Using this method, the output layer weights can be obtained without iteration. When the constraints are added, the problem of solving the output layer weights becomes a constrained optimization problem.

Among them, the training process of the online sequence overrun learning machine includes an initial ELM batch learning process and a continuous sequential learning process, as follows:

In the initialization phase, given the initial training subset

Among them, N ₁ is the newly arrived data, by

The calculated initial output weight is

among them,

Whenever a new training sample is obtained

When

Recursively calculate output weights. among them,

As an example, when calculating the variable impedance parameter of the teaching motion based on the interaction force data in the teaching data, the variable stiffness parameter and the variable damping parameter may be calculated based on the interaction force data.

Specifically, when calculating the variable stiffness parameter, let

Represents the collected interaction force (F) and corresponding time (q) information, where is the number of disturbance data samples obtained. The variable stiffness parameter at time q is calculated from the force information in the time window [q-(w-1), q]. The length of the sliding time window is w, and the upper and lower bounds of the data points in the window are represented by L _q and U _q ,

The number of data points in the window at time q is W _q =U _q -L _q +1, and the covariance matrix corresponding to the force data in the window is

among them,

Because the covariance matrix Σ _q is symmetric and positive definite, it can be decomposed into the following form Σ _q = PΛP- ¹ , where Λ is the eigenvalue

Diagonal array. The stiffness matrix K _q is

among them,

And eigenvalues

Proportional to the expression

As the teaching progresses, the interactive force data will be continuously collected, and the new data will be sorted according to the time information and the values in the window will be taken to solve the stiffness. For example, when the data at time q+1 enters, the online update of the covariance is

among them,

Specifically, when calculating the variable damping parameter, since the damping ratio is constant, the square root of the damping and the stiffness is linear, so it can be based on the formula

To calculate the variable damping parameter B. Among them, γ is a constant greater than 0.

In the embodiment of the present invention, in order to ensure the stability of the learned model, preferably, when calculating the variable impedance parameter of the teaching motion based on the interaction force data, the preset stability constraints and the interaction force data can be used to predict The variable impedance model is trained to obtain the variable impedance parameters of the teaching movement, and the variable impedance model is updated according to the variable impedance parameters, so as to ensure the stability of the variable impedance control and avoid excessive robot interaction force that may cause injury. Happening.

In step S103, the operation is controlled according to the equation of motion and the variable impedance parameter.

In the embodiment of the present invention, after obtaining the motion equation and the variable impedance parameter, the operation can be controlled according to the motion equation and the variable impedance parameter, thereby controlling the robot to reproduce the movement trajectory and interactive force of the teaching movement.

In the embodiment of the present invention, after the training of the preset neural network model (for example, online sequence overrun learning machine) and variable impedance model is completed, the trained neural network model (for example, online sequence overrun learning machine) can be used And variable impedance models to control the trajectory and interaction force of the robot to reproduce the teaching movement.

As an example, FIG. 4 shows an exemplary diagram of teaching learning and reproduction of robot compliance control. As shown in FIG. 4, the instructor grasps the robot with one hand for teaching, and the robot collects Track information

And force information F _q , then according to the trajectory information

Perform motion learning to obtain f(·), and learn variable stiffness parameters and variable damping parameters according to the force information F _q to get {B _q ,K _q }, and finally generate motion according to f(·) and according to {B _q ,K _q }Variable impedance control to control the trajectory and interaction force of the robot to reproduce the teaching movement.

In the embodiment of the present invention, by acquiring the teaching data of the teaching movement, the motion equation of the teaching movement is calculated according to the movement data in the teaching data, and at the same time, the variation of the teaching movement is calculated according to the interaction force data in the teaching data Impedance parameters, according to the motion equation and variable impedance parameter control operation, thereby reducing the manual programming in the robot compliance control process, lowering the threshold for robot use, improving the flexibility and accuracy of robot control, and thus improving the robot's universal Ability, degree of intelligence and control effect.

Example 2:

FIG. 5 shows the structure of the robot compliance control device provided in Embodiment 2 of the present invention. For ease of explanation, only parts related to the embodiment of the present invention are shown, including: a data acquisition unit 51, a parameter calculation unit 52 and Operation control unit 53.

The data acquiring unit 51 is configured to acquire teaching data of the teaching movement, wherein the teaching data includes at least the movement data and the interaction force data of the teaching movement.

In the embodiment of the present invention, the learning of the teaching movement may include action learning and force learning (ie, variable stiffness parameter and variable damping parameter learning).

Therefore, preferably, the data acquisition unit 51 may include:

The first acquiring unit is used to acquire position data, interaction force data and time data related to the teaching movement; and

The first calculation unit is used to calculate the motion data according to the position data and the time data.

Specifically, the speed data related to the teaching movement is calculated based on the position data and the time data, thereby obtaining the movement data of the teaching movement.

Preferably, the data acquisition unit 51 may further include:

The second acquisition unit is used to acquire angle data, interaction force data and time data related to the teaching movement; and

The second calculation unit is used to calculate the motion data according to the angle data and the time data.

Specifically, the angular acceleration data related to the teaching movement may be calculated according to the angle data and the time data, thereby obtaining the movement data of the teaching movement.

The parameter calculation unit 52 is configured to calculate the motion equation of the teaching motion based on the motion data in the teaching data, and at the same time calculate the variable impedance parameter of the teaching motion based on the interaction force data in the teaching data, where the variable impedance parameter includes at least Variable stiffness parameters and variable damping parameters.

In the embodiment of the present invention, the trajectory and the force are simultaneously learned, thereby improving the learning effect, and thereby improving the accuracy of the control result.

Preferably, the parameter calculation unit 52 may include:

The first training unit is used to train the preset neural network model using motion data to obtain the motion equation of the teaching movement, and update the neural network model online according to the motion equation, thereby improving the calculation efficiency of the motion equation and facilitating movement The subsequent use of equations, and to meet the needs of real-time online learning, and thus improve the learning effect.

Among them, preferably, the model training unit may include:

The incremental learning unit is used to incrementally learn the motion data in a one-by-one or block-by-block manner to obtain the motion equation of the teaching motion, thereby improving the accuracy of the motion equation and thereby improving the learning effect.

Among them, preferably, the neural network model is an online sequence overrun learning machine.

Preferably, the parameter calculation unit 52 may include:

The second training unit is used to train the preset variable impedance model according to the preset stability constraints and interaction force data to obtain the variable impedance parameter of the teaching movement, and update the variable impedance model according to the variable impedance parameter, Therefore, the stability of the variable impedance control is ensured, and the situation that the interaction force of the robot is too large to cause injury is avoided.

The operation control unit 53 is used to control the operation according to the equation of motion and the variable impedance parameter.

In the embodiment of the present invention, the teaching data of the teaching movement is acquired by the data acquiring unit 51, the motion equation of the teaching movement is calculated according to the movement data in the teaching data by the parameter calculating unit 52, and at the same time according to the teaching data The interactive force data calculates the variable impedance parameters of the teaching movement, and the operation is controlled by the operation control unit 53 according to the motion equation and the variable impedance parameters, thereby reducing the manual programming in the robot compliance control process, lowering the robot's use threshold, and improving the robot The flexibility and accuracy of the control, thereby improving the robot's generalization ability, intelligence and control effect.

In the embodiment of the present invention, each unit of the robot compliance control device may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into one software and hardware unit, which is not limited here. this invention.

Example three:

FIG. 6 shows the structure of the computing device provided in Embodiment 4 of the present invention. For ease of description, only parts related to the embodiment of the present invention are shown.

The computing device 6 of the embodiment of the present invention includes a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60. When the processor 60 executes the computer program 62, the steps in the above embodiments of the robot compliance control method are implemented, for example, steps S101 to S103 shown in FIG. 1. Alternatively, when the processor 60 executes the computer program 62, the functions of the units in the above device embodiments are realized, for example, the functions of the units 51 to 53 shown in FIG.

In the embodiment of the present invention, when the processor 60 executes the computer program 62 to realize the steps in the above embodiments of the robot compliance control method, the teaching data of the teaching motion is acquired, and the teaching data is calculated according to the motion data in the teaching data Teaching the motion equation of motion, and at the same time calculating the variable impedance parameters of the teaching motion based on the interactive force data in the teaching data, and controlling the operation according to the motion equations and variable impedance parameters, thereby reducing the manual programming and reducing the robot's compliance control process. The use threshold of the robot is improved, and the flexibility and accuracy of the robot control are improved, thereby improving the robot's generalization ability, intelligence, and control effect.

For the steps implemented by the processor 60 in the computing device 6 when executing the computer program 62, reference may be made to the description of the method in Embodiment 1, and details are not described herein again.

Example 5:

In an embodiment of the present invention, a computer-readable storage medium is provided, and the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the embodiments of the foregoing robot compliance control methods are implemented. For example, steps S101 to S103 shown in FIG. 1. Alternatively, when the computer program is executed by the processor, the functions of the units in the foregoing device embodiments are realized, for example, the functions of the units 51 to 53 shown in FIG. 5.

In the embodiment of the present invention, the teaching data of the teaching motion is acquired, the motion equation of the teaching motion is calculated according to the motion data in the teaching data, and the variable impedance of the teaching motion is calculated according to the interaction force data in the teaching data The parameters control the operation according to the equations of motion and variable impedance parameters, thereby reducing manual programming during robot compliance control, lowering the threshold for robot use, improving the flexibility and accuracy of robot control, and thus improving the generalization of the robot Ability, degree of intelligence and control effect. The robot compliance control method implemented when the computer program is executed by the processor may further refer to the description of the steps in the foregoing method embodiments, and details are not described herein again.

The computer-readable storage medium in the embodiments of the present invention may include any entity or device capable of carrying computer program code, and a recording medium, such as ROM/RAM, magnetic disk, optical disk, flash memory, and other memories.

The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modification, equivalent replacement and improvement made within the spirit and principle of the present invention should be included in the protection of the present invention Within range.

Claims

A robot compliance control method, characterized in that the method includes the following steps:

Acquiring teaching data of a teaching movement, wherein the teaching data includes at least the movement data and interaction force data of the teaching movement;

Calculating the motion equation of the teaching motion based on the motion data in the teaching data, and simultaneously calculating the variable impedance parameter of the teaching motion based on the interaction force data in the teaching data, wherein the variable impedance The parameters include at least variable stiffness parameters and variable damping parameters;

The operation is controlled according to the equation of motion and the variable impedance parameter.
The method of claim 1, wherein the step of acquiring teaching data of the teaching movement includes:

Obtain position data, interaction force data and time data related to teaching movement;

The motion data is calculated based on the position data and the time data, wherein the motion data includes speed data.
The method according to claim 1, wherein the step of calculating the motion equation of the teaching motion based on the motion data in the teaching data includes:

Training the preset neural network model using the motion data to obtain the motion equation of the teaching movement, and updating the neural network model online according to the motion equation,

Wherein, the step of training the preset neural network model using the motion data includes:

Perform incremental learning on the motion data in a one-by-one or block-by-block manner to obtain the motion equation of the teaching motion.
The method of claim 4, wherein the neural network model includes an online sequence overrun learning machine.
The method of claim 1, wherein the step of calculating the variable impedance parameter of the teaching motion based on the interaction force data in the teaching data includes:

Train a preset variable impedance model according to preset stability constraints and the interaction force data to obtain a variable impedance parameter of the teaching movement, and update the variable impedance model according to the variable impedance parameter .
A robot compliance control device, characterized in that the device includes:

A data acquisition unit for acquiring teaching data of the teaching movement, wherein the teaching data includes at least the movement data and the interaction force data of the teaching movement;

A parameter calculation unit, configured to calculate the motion equation of the teaching motion based on the motion data in the teaching data, and at the same time calculate the variable impedance parameter of the teaching motion based on the interaction force data in the teaching data, Wherein, the variable impedance parameter includes at least a variable stiffness parameter and a variable damping parameter; and

The operation control unit is used for controlling operation according to the motion equation and the variable impedance parameter.
The apparatus of claim 6, wherein the data acquisition unit comprises:

The first acquiring unit is used to acquire position data, interaction force data and time data related to the teaching movement; and

The first calculation unit is configured to calculate the motion data according to the position data and the time data, wherein the motion data includes speed data.
The apparatus of claim 6, wherein the parameter calculation unit comprises:

The first training unit is used to train the preset neural network model using the motion data to obtain the motion equation of the teaching movement, and update the neural network model online according to the motion equation,

Wherein, the model training unit includes:

An incremental learning unit is used to incrementally learn the motion data in a one-by-one or block-by-block manner to obtain the motion equation of the teaching motion.
The apparatus of claim 8, wherein the neural network model includes an online sequence overrun learning machine.
The apparatus of claim 6, wherein the parameter calculation unit comprises:

The second training unit is configured to train the preset variable impedance model according to the preset stability constraint conditions and the interaction force data, to obtain the variable impedance parameter of the teaching movement, and according to the variable impedance parameter The variable impedance model is updated.
A computing device, including 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 is implemented as claimed in claims 1 to 5 The steps of any of the methods described.
A computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.