Compensation method for friction force error of robot arm

Abstract

The invention provides a robot arm friction force error compensation method. The method comprises the steps that joint friction force compensation data of a robot and standard compensation friction force corresponding to the joint friction force compensation data are obtained; and performing prediction based on the joint friction force compensation data by using a Kolmogorov-Arnod layer, a Kolmogorov-Arnod network and a long-short term memory network to obtain predicted compensation friction force. Adjusting the parameters of the pre-training friction force compensation model based on the difference between the standard compensation friction force and the predicted compensation friction force; and continuously utilizing the joint friction force compensation data of the robot and the standard compensation friction force corresponding to the joint friction force compensation data to train the pre-training friction force compensation model to obtain a manipulator friction force compensation model. And the friction force error of the manipulator is compensated through the manipulator friction force compensation model. According to the method provided by the invention, the friction force error of the robot arm can be more accurately compensated.

Description

Technical Field

[0001] The present application relates to the field of robotics technology, and in particular to a method for compensating friction errors of a robot arm. Background Art

[0002] With the continuous development of science and technology, robots are increasingly being used in our daily lives. The robotic arm skeleton device is a precision device used to help patients with hand function recovery training, and is widely used in the fields of sports rehabilitation and neurological rehabilitation. Patients with hand dysfunction caused by nerve damage or other diseases need to restore hand function through repetitive and delicate operation training. However, in actual applications, the joints and sliding parts of the robotic arm skeleton are affected by friction, resulting in a decrease in motion control accuracy.

[0003] The existing technology usually relies on the robot joint friction model to supplement the manipulator friction. The robot joint friction model is to add a simplified friction model (such as Coulomb friction, viscous resistance model, Stribeck model, Karnopp model, etc.) to the robot dynamics model, and identify the model parameters by combining the parameter identification method. However, the existing technology relies on the robot joint friction model to compensate for the error of the manipulator friction, which often has low accuracy.

[0004] Therefore, how to improve the compensation accuracy of the friction error of the robot arm has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0005] Based on the above problems, the present application provides a method for compensating a friction error of a robot arm to improve the compensation accuracy of the friction error of the robot arm.

[0006] The present application provides a method for compensating a friction error of a robot arm, the method comprising the following steps: Acquire joint friction force compensation data of the robot and standard compensation friction force corresponding to the joint friction force compensation data, wherein the joint friction force compensation data includes joint curvature, joint output power, joint mass and joint length; Performing feature extraction on the joint friction force compensation data to obtain a feature extraction result; Performing layer normalization processing on the feature extraction result using a pre-trained friction compensation model to obtain a first normalized result; Amplifying the first normalized result by a preset multiple to obtain an amplified result; Processing the amplified result through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector and a reference vector; determining a value vector based on the amplification result; Processing the query vector, the key vector, and the value vector through a long short-term memory network of the pre-trained friction compensation model under the guidance of the amplification result to obtain a first processing result; Performing group normalization processing on the first processing result using a pre-trained friction compensation model to obtain a second normalized result; Performing feature fusion on the second normalization result, the reference vector, and the magnification result to obtain a feature fusion result; Reducing the feature fusion result according to a preset ratio to obtain a reduced result; Processing the reduced result through a Kolmogorov-Arnold layer to obtain a second processed result; Performing channel normalization processing on the second processing result using a pre-trained friction compensation model to obtain a predicted compensation friction force; The manipulator friction compensation model is obtained by adjusting the parameters of the pre-trained friction compensation model based on the difference between the standard compensation friction force and the predicted compensation friction force, and continuously training the pre-trained friction compensation model using the joint friction compensation data of the robot and the standard compensation friction force corresponding to the joint friction compensation data; The friction error of the robot is compensated by the robot friction compensation model.

[0007] In a possible implementation, the Kolmogorov-Arnold network includes a 1x4 convolution, a 1x1 convolution, a 4x1 convolution, a Kolmogorov-Arnold layer, a summation layer, and an activation function, and the amplification result is processed through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector, and a reference vector, including: The amplified result is respectively subjected to the 1x4 convolution, the 1x1 convolution and the 4x1 convolution to obtain a first convolution result, a second convolution result and a third convolution result; Summing the first convolution result, the second convolution result and the third convolution result through the summation layer to obtain a summation result; Inputting the summation result into the Kolmogorov-Arnold layer to obtain a nonlinear learning result; Fusing the nonlinear learning result and the amplification result to obtain a nonlinear fusion result; The nonlinear fusion result is obtained by using an activation function to obtain the query vector, the key vector and the reference vector.

[0008] In a possible implementation, the activation function is expressed by the following formula: LeakySigmoid(x)=x(Sigmoid(x)+αmax(0,-x)); Wherein, LeakySigmoid(x) represents the activation function, α represents a constant, and x represents the input of the activation function. .

[0009] In a possible implementation, the parameters of the long short-term memory network are expressed by the following formula: ; Where t represents time, represents the memory state at time t, represents the first intermediate quantity at time t, represents the memory state at time t-1, represents the second intermediate quantity at time t, represents the value vector at time t, represents the key vector at time t; ; Among them, t represents time, n represents time series, represents the time series corresponding to time t, represents the first intermediate quantity at time t, represents the time series at time t-1, represents the second intermediate quantity at time t, represents the key vector at time t; ; Where t represents time, represents the measured value of the state quantity at time t, represents the state quantity at time t, represents the third intermediate quantity, Related to the query vector at time t.

[0010] In a possible implementation, the error value of the state quantity at time t is expressed by the following formula: ; Where t represents time, represents the measured value of the state quantity at time t, represents the memory state at time t, represents the query vector at time t, Represents the time series corresponding to time t.

[0011] In a possible implementation, the query vector, key vector and reference vector are expressed by the following formula: ; Where t represents time, represents the query vector at time t, represents the input value at time t, represents the first weight, represents the first component; ; Where t represents time, Indicates that the dimension of the query vector q is d, represents the key vector at time t, represents the input value at time t, represents the second weight, represents the second component; ; Where t represents time, represents the value vector at time t, represents the input value at time t, represents the third weight, Represents the third component.

[0012] In a possible implementation, the first intermediate quantity, the second intermediate quantity, and the third intermediate quantity are expressed by the following formula: ; Where t represents time, represents the first intermediate quantity at time t, represents the measured value of the first intermediate quantity at time t; ; in, represents the first parameter corresponding to the first intermediate quantity, represents the second parameter corresponding to the first intermediate quantity, represents the parameter corresponding to time t; ; Where t represents time, represents the second intermediate quantity at time t, represents the measured value of the second intermediate quantity at time t; ; in, represents the first parameter corresponding to the second intermediate quantity, represents the second parameter corresponding to the second intermediate quantity, represents the parameter corresponding to time t; ; Where t represents time, represents the third intermediate quantity at time t, represents the measured value of the third intermediate quantity at time t; ; in, represents the first parameter corresponding to the third intermediate quantity, represents the second parameter corresponding to the third intermediate quantity, Represents the input value at time t.

[0013] The present application also provides a compensation device for friction error of a robot arm, the device comprising: An acquisition module, used for acquiring joint friction force compensation data of the robot and a standard compensation friction force corresponding to the joint friction force compensation data, wherein the joint friction force compensation data includes joint curvature, joint output power, joint mass and joint length; A feature extraction module, used for performing feature extraction on the joint friction force compensation data to obtain a feature extraction result; A first normalization module, used to perform layer normalization processing on the feature extraction result using a pre-trained friction compensation model to obtain a first normalized result; An upper projection module, used for magnifying the first normalized result by a preset multiple to obtain a magnified result; A nonlinear learning module, used for processing the amplification result through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector and a reference vector; A determination module, configured to determine a value vector based on the amplification result; A first processing module, configured to process the query vector, the key vector and the value vector through a long short-term memory network of the pre-trained friction compensation model under the guidance of the amplification result to obtain a first processing result; A second normalization module, used for performing group normalization processing on the first processing result using a pre-trained friction compensation model to obtain a second normalized result; A fusion module, used for performing feature fusion on the second normalization result, the reference vector, and the amplification result to obtain a feature fusion result; A lower projection module, used for reducing the feature fusion result according to a preset ratio to obtain a reduced result; A second processing module, used for processing the reduction result through a Kolmogorov-Arnold layer to obtain a second processing result; a third normalization module, configured to perform channel normalization processing on the second processing result using a pre-trained friction compensation model to obtain a predicted compensation friction force; A training module, configured to adjust the parameters of the pre-trained friction compensation model based on the difference between the standard compensation friction force and the predicted compensation friction force, and continuously train the pre-trained friction compensation model using the joint friction compensation data of the robot and the standard compensation friction force corresponding to the joint friction compensation data to obtain a manipulator friction compensation model; The compensation module is used to compensate the friction error of the robot arm through the robot arm friction compensation model.

[0014] The present application also provides an electronic device, the electronic device comprising a processor and a memory: The memory is used to store a computer program and transmit the computer program to the processor; The processor is used to execute the steps of the above-mentioned method for compensating the friction error of the robot according to the instructions in the computer program.

[0015] The present application also provides a computer-readable storage medium, which is used to store a computer program. When the computer program is executed by an electronic device, the steps of the method for compensating the friction error of the robot arm are implemented.

[0016] Compared with the prior art, this application has the following beneficial effects: The method provided by the present application can capture the changing law of friction in a higher dimension, perform accurate modeling, and provide a more accurate friction compensation effect by combining the Kolmogorov-Arnold network and the long short-term memory network in deep learning. The present invention extracts multiple key features including joint curvature, output power, joint mass and joint length, combines the feature fusion technology in deep learning, fully explores the potential relationship in the data, and improves the accuracy of friction compensation. In addition, through the processing of normalization and feature fusion, the difference between different data dimensions is effectively reduced, and the robustness of the model in different environments and working conditions is guaranteed. Under the condition of unknown contact surface, the present application combines the Kolmogorov-Arnold network and the long short-term memory network, considers the contact surface conditions such as non-local memory effect, critical friction and viscous friction, introduces non-local memory effect, critical friction, and viscous friction nonlinear activation function, based on data-driven identification of the optimal estimated state of the contact interface, considers the situation of the contact interface to select a supplementary strategy, can maximize the friction state of the contact surface, so as to more accurately compensate for the friction error of the manipulator. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0018] Figure 1 A flowchart of a method for compensating a friction error of a robot provided in an embodiment of the present application; Figure 2 A schematic diagram of a robot arm provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a Kolmogorov-Arnold network provided in an embodiment of the present application; Figure 4 A schematic diagram of the structure of a long short-term memory network provided in an embodiment of the present application; Figure 5 A schematic diagram of the model structure of a pre-trained friction compensation model provided in an embodiment of the present application; Figure 6 A schematic diagram of the structure of a device for compensating friction error of a robot arm provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] As described above, the existing technology usually relies on the robot joint friction model to supplement the manipulator friction. The robot joint friction model is to add a simplified friction model (such as Coulomb friction, viscous resistance model, Stribeck model, Karnopp model, etc.) to the robot dynamics model, and identify the model parameters by combining the parameter identification method. However, the existing technology relies on the robot joint friction model to compensate for the error of the manipulator friction force, which often has low accuracy.

[0020] After research, it was found that due to the particularity of the robot arm, when the robot arm is not in use, the contact interface between the user and the robot arm is unknown. The robot joint friction model in the related technology is not conducive to the selection of the basic model and its combination when the contact interface is unknown, and the friction force identification accuracy is low. In addition, due to the particularity of the robot arm, there are more nonlinear friction forces during the use of the robot arm. The method of combining machine learning for dynamic modeling in the related technology can improve the nonlinear fitting effect through multi-layer neural networks, but for the robot arm model with multiple moving joints, the motion space state is complex and it is impossible to traverse all states of the robot arm's motion process, which is not conducive to realizing the dynamic control of the robot in any state. Therefore, the error compensation of the friction force of the robot arm in the related technology is often of low accuracy.

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] It is understandable that the method provided in the present application can be applied to a processing device, which is a processing device that can obtain the joint friction compensation data of the robot, such as a terminal device or a server that can obtain the joint friction compensation data of the robot. The method provided in the present application can be executed independently by a terminal device or a server, and can also be applied to a network scenario in which a terminal device and a server communicate, and is executed by the cooperation of a terminal device and a server. Among them, the terminal device can be a computer, a mobile phone and other devices. The server can be understood as an application server or a Web server. In actual deployment, the server can be an independent server or a cluster server.

[0023] Figure 1 A flowchart of a method for compensating a friction error of a robot provided in the present application, the method comprising the following steps: S101: Acquire joint friction force compensation data of the robot and standard compensation friction force corresponding to the joint friction force compensation data.

[0024] The processing device obtains joint friction compensation data of the robot and standard compensation friction corresponding to the joint friction compensation data. The joint friction compensation data includes joint bending, joint output power, joint mass and joint length; Figure 2 A schematic diagram of a robot arm provided in an embodiment of the present application, Figure 2 Joint 1 and joint N are marked in the figure. Figure 2 The robotic hand on display includes multiple joints.

[0025] In a possible implementation, the processing device may obtain the joint curvatures of the N joints, namely, joint 1 to joint N. Taking joint 1 as an example, joint 1 may vary from -5° to 90°, so the joint curvature of joint 1 is (-5°, 90°).

[0026] The processing device can collect the output power of the motor corresponding to each joint of the robot through the power supply system of the robot. Taking joint 1 as an example, the kinetic energy taken by the processing device when moving only joint 1 while ensuring that other joints do not move is .therefore, It can be thought of as a series of points where the value changes as the joint changes from -5° to 90°.

[0027] The processing device can obtain the joint mass of the joint rotation part of the mechanical glove involved in the joint rotation through the mechanical glove properties, and can also obtain the joint length in the mechanical glove.

[0028] S102: Perform feature extraction on the joint friction force compensation data to obtain a feature extraction result.

[0029] The processing device performs feature extraction on the joint friction force compensation data to obtain a feature extraction result.

[0030] Feature extraction can be performed through the feature extraction layer of a pre-trained friction compensation model or through a separate feature extraction model.

[0031] S103: performing layer normalization processing on the feature extraction result using the pre-trained friction compensation model to obtain a first normalized result.

[0032] The processing device performs layer normalization processing on the feature extraction result using a pre-trained friction compensation model to obtain a first normalized result.

[0033] Layer normalization normalizes the input data by calculating the mean and variance of all feature values ​​in a single sample, which can be directly applied to recurrent neural networks. After normalization, two learnable parameters, adaptive bias and gain, are also provided to enhance the expressiveness of the model. Layer normalization performs exactly the same calculations during training and testing.

[0034] S104: Amplify the first normalized result by a preset multiple to obtain an amplified result.

[0035] The processing device amplifies the first normalized result by a preset multiple to obtain an amplified result.

[0036] Since the friction data of the robot hand is small, in order to facilitate model training and make the subsequent use process more accurate, the first normalized result can be magnified by a preset multiple to obtain an amplified result. In a possible implementation, the preset multiple can be twice.

[0037] S105: Processing the amplified result through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector and a reference vector.

[0038] The processing device processes the amplified result through a Kolmogorov-Arnold network of a pre-trained friction compensation model to obtain a query vector, a key vector, and a reference vector.

[0039] In one possible implementation, the Kolmogorov-Arnold network includes a 1x4 convolution, a 1x1 convolution, a 4x1 convolution, a Kolmogorov-Arnold layer, a summation layer, and an activation function.

[0040] The Kolmogorov-Arnold layer applies a learnable nonlinear activation function to each input element.

[0041] The processing device may respectively perform 1x4 convolution, 1x1 convolution, and 4x1 convolution on the amplified result to obtain a first convolution result, a second convolution result, and a third convolution result.

[0042] The first convolution result, the second convolution result and the third convolution result are summed up through a summation layer to obtain a summation result.

[0043] The summation result is input into the Kolmogorov-Arnold layer to obtain the nonlinear learning result.

[0044] The nonlinear learning result and the amplification result are fused to obtain the nonlinear fusion result.

[0045] The nonlinear fusion results are obtained by using the activation function to obtain the query vector, key vector and reference vector.

[0046] In one possible implementation, the activation function is expressed as follows: LeakySigmoid(x)=x(Sigmoid(x)+αmax(0,-x)); Wherein, LeakySigmoid(x) represents the activation function, α represents a constant, and x represents the input of the activation function. .

[0047] Figure 3 A schematic diagram of the structure of a Kolmogorov-Arnold network provided in an embodiment of the present application.

[0048] S106: Determine a value vector based on the amplification result.

[0049] The processing device determines a value vector based on the upscaling result.

[0050] S107: The query vector, the key vector, and the value vector are processed by a long short-term memory network of a pre-trained friction compensation model under the guidance of the amplification result to obtain a first processing result.

[0051] The processing device processes the query vector, the key vector and the value vector through a long short-term memory network of a pre-trained friction compensation model under the guidance of the amplification result to obtain a first processing result.

[0052] Long Short-Term Memory (LSTM) is a special type of recurrent neural network (RNN) that aims to solve the gradient vanishing and gradient exploding problems encountered by traditional RNNs when processing long sequence data. LSTM effectively captures and retains long-term dependency information by introducing memory units and gating mechanisms. When using the long short-term memory network for processing, the attention mechanism is used. The query vector (Query), key vector (Key) and value vector (Value) are processed by the long short-term memory network under the guidance of the amplification result to obtain the first processing result.

[0053] In one possible implementation, the parameters of the long short-term memory network are expressed by the following formula: ; Where t represents time, represents the memory state at time t, represents the first intermediate quantity at time t, represents the memory state at time t-1, represents the second intermediate quantity at time t, represents the value vector at time t, represents the key vector at time t; ; Among them, t represents time, n represents time series, represents the time series corresponding to time t, represents the first intermediate quantity at time t, represents the time series at time t-1, represents the second intermediate quantity at time t, represents the key vector at time t; ; Where t represents time, represents the measured value of the state quantity at time t, represents the state quantity at time t, represents the third intermediate quantity, Related to the query vector at time t.

[0054] In a possible implementation, the error value of the state quantity at time t is expressed by the following formula: ; Where t represents time, represents the measured value of the state quantity at time t, represents the memory state at time t, represents the query vector at time t, Represents the time series corresponding to time t.

[0055] In one possible implementation, the query vector, key vector, and reference vector are expressed by the following formula: ; Where t represents time, represents the query vector at time t, represents the input value at time t, represents the first weight, represents the first component; ; Where t represents time, Indicates that the dimension of the query vector q is d, represents the key vector at time t, represents the input value at time t, represents the second weight, represents the second component; ; Where t represents time, represents the value vector at time t, represents the input value at time t, represents the third weight, Represents the third component.

[0056] In a possible implementation, the first intermediate quantity, the second intermediate quantity and the third intermediate quantity are expressed by the following formula: ; Where t represents time, represents the first intermediate quantity at time t, represents the measured value of the first intermediate quantity at time t; ; in, represents the first parameter corresponding to the first intermediate quantity, represents the second parameter corresponding to the first intermediate quantity, represents the parameter corresponding to time t; ; Where t represents time, represents the second intermediate quantity at time t, represents the measured value of the second intermediate quantity at time t; ; in, represents the first parameter corresponding to the second intermediate quantity, represents the second parameter corresponding to the second intermediate quantity, represents the parameter corresponding to time t; ; Where t represents time, represents the third intermediate quantity at time t, represents the measured value of the third intermediate quantity at time t; ; in, represents the first parameter corresponding to the third intermediate quantity, represents the second parameter corresponding to the third intermediate quantity, Represents the input value at time t.

[0057] Figure 4 A schematic diagram of the structure of a long short-term memory network provided in an embodiment of the present application.

[0058] S108: Performing group normalization processing on the first processing result using a pre-trained friction compensation model to obtain a second normalized result.

[0059] The processing device performs group normalization processing on the first processing result using a pre-trained friction compensation model to obtain a second normalized result.

[0060] The main idea of ​​Group Normalization (GN) is to divide the channels into several groups and then normalize the channels in each group. This can reduce the amount of model calculation while maintaining the normalization effect.

[0061] S109: Perform feature fusion on the second normalization result, the reference vector, and the magnification result to obtain a feature fusion result.

[0062] The processing device performs feature fusion on the second normalization result, the reference vector, and the amplification result to obtain a feature fusion result.

[0063] S110: reducing the feature fusion result according to a preset ratio to obtain a reduced result.

[0064] The processing device reduces the feature fusion result according to a preset ratio to obtain a reduced result.

[0065] In a possible implementation, the preset ratio may correspond to the preset multiple mentioned above. For example, if the preset multiple is 2, the preset ratio may be 1 / 2.

[0066] S111: Processing the reduced result through the Kolmogorov-Arnold layer to obtain a second processing result.

[0067] The processing device processes the reduction result through the Kolmogorov-Arnold layer to obtain a second processing result. The Kolmogorov-Arnold layer also performs nonlinear prediction when the contact surface is unknown.

[0068] S112: performing channel normalization processing on the second processing result using a pre-trained friction compensation model to obtain a predicted compensation friction force.

[0069] The processing device performs channel normalization processing on the second processing result using a pre-trained friction compensation model to obtain a predicted compensation friction force.

[0070] S113: The manipulator friction compensation model is obtained by adjusting the parameters of the pre-trained friction compensation model based on the difference between the standard compensation friction force and the predicted compensation friction force, and continuously training the pre-trained friction compensation model using the robot's joint friction compensation data and the standard compensation friction force corresponding to the joint friction compensation data.

[0071] The processing device adjusts the parameters of the pre-trained friction compensation model based on the difference between the standard compensation friction and the predicted compensation friction, and continuously trains the pre-trained friction compensation model using the robot's joint friction compensation data and the standard compensation friction corresponding to the joint friction compensation data to obtain a manipulator friction compensation model.

[0072] In a possible implementation, the parameters of the pre-trained friction compensation model that can be adjusted may include the above-mentioned wait.

[0073] S114: Compensating the friction error of the robot arm by using the robot arm friction compensation model.

[0074] The processing device compensates for the friction error of the robot arm through the robot arm friction compensation model.

[0075] Figure 5 A schematic diagram of the model structure of a pre-trained friction compensation model provided in an embodiment of the present application.

[0076] The method provided by the present application can capture the changing law of friction in a higher dimension, perform accurate modeling, and provide a more accurate friction compensation effect by combining the Kolmogorov-Arnold network and the long short-term memory network in deep learning. The present invention extracts multiple key features including joint curvature, output power, joint mass and joint length, combines the feature fusion technology in deep learning, fully explores the potential relationship in the data, and improves the accuracy of friction compensation. In addition, through the processing of normalization and feature fusion, the difference between different data dimensions is effectively reduced, and the robustness of the model in different environments and working conditions is guaranteed. Under the condition of unknown contact surface, the present application combines the Kolmogorov-Arnold network and the long short-term memory network, considers the contact surface conditions such as non-local memory effect, critical friction and viscous friction, introduces non-local memory effect, critical friction, and viscous friction nonlinear activation function, based on data-driven identification of the optimal estimated state of the contact interface, considers the situation of the contact interface to select a supplementary strategy, can maximize the friction state of the contact surface, so as to more accurately compensate for the friction error of the manipulator.

[0077] The present application also provides a Figure 6 The schematic diagram of the structure of the compensation device for the friction error of the robot arm is shown, and the compensation device 600 for the friction error of the robot arm includes: An acquisition module 601 is used to acquire joint friction compensation data of the robot and a standard compensation friction corresponding to the joint friction compensation data, wherein the joint friction compensation data includes joint curvature, joint output power, joint mass and joint length; A feature extraction module 602 is used to extract features from the joint friction force compensation data to obtain feature extraction results; A first normalization module 603, used to perform layer normalization processing on the feature extraction result using a pre-trained friction compensation model to obtain a first normalized result; An upper projection module 604 is used to magnify the first normalized result by a preset multiple to obtain a magnified result; A nonlinear learning module 605, configured to process the amplification result through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector and a reference vector; A determination module 606, configured to determine a value vector based on the amplification result; A first processing module 607 is used to process the query vector, the key vector and the value vector through the long short-term memory network of the pre-trained friction compensation model under the guidance of the amplification result to obtain a first processing result; A second normalization module 608, configured to perform group normalization processing on the first processing result using a pre-trained friction compensation model to obtain a second normalized result; A fusion module 609 is used to perform feature fusion on the second normalization result, the reference vector, and the magnification result to obtain a feature fusion result; A lower projection module 610 is used to reduce the feature fusion result according to a preset ratio to obtain a reduced result; A second processing module 611 is used to process the reduction result through a Kolmogorov-Arnold layer to obtain a second processing result; A third normalization module 612, configured to perform channel normalization processing on the second processing result using a pre-trained friction compensation model to obtain a predicted compensation friction force; A training module 613 is used to adjust the parameters of the pre-trained friction compensation model based on the difference between the standard compensation friction force and the predicted compensation friction force, and continuously train the pre-trained friction compensation model using the joint friction compensation data of the robot and the standard compensation friction force corresponding to the joint friction compensation data to obtain a manipulator friction compensation model; The compensation module 614 is used to compensate the friction error of the robot arm by using the robot arm friction compensation model.

[0078] The device provided by the present application can capture the changing law of friction in a higher dimension, perform accurate modeling, and provide a more accurate friction compensation effect by combining the Kolmogorov-Arnold network and the long short-term memory network in deep learning. The present invention extracts multiple key features including joint curvature, output power, joint mass and joint length, combines the feature fusion technology in deep learning, fully explores the potential relationship in the data, and improves the accuracy of friction compensation. In addition, through the processing of normalization and feature fusion, the difference between different data dimensions is effectively reduced, and the robustness of the model in different environments and working conditions is guaranteed. Under the condition of unknown contact surface, the present application combines the Kolmogorov-Arnold network and the long short-term memory network, considers the contact surface conditions such as non-local memory effect, critical friction and viscous friction, introduces non-local memory effect, critical friction, and viscous friction nonlinear activation function, based on data-driven identification of the optimal estimated state of the contact interface, considers the situation of the contact interface to select a supplementary strategy, can maximize the friction state of the contact surface, thereby more accurately compensating the friction error of the robot.

[0079] An embodiment of the present application also provides a compensation device for the friction error of a robot arm, wherein the device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device performs the steps of the method for compensating the friction error of a robot arm described in any embodiment of the present application.

[0080] In practical applications, the computer-readable storage medium may be any combination of one or more computer-readable media, which may be a computer-readable signal medium or a computer-readable storage medium.

[0081] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0082] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0083] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0084] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0085] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0086] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for compensating friction error of a robot arm, characterized in that: include: Acquire joint friction force compensation data of the robot and standard compensation friction force corresponding to the joint friction force compensation data, wherein the joint friction force compensation data includes joint curvature, joint output power, joint mass and joint length; Performing feature extraction on the joint friction force compensation data to obtain a feature extraction result; Performing layer normalization processing on the feature extraction result using a pre-trained friction compensation model to obtain a first normalized result; Amplifying the first normalized result by a preset multiple to obtain an amplified result; Processing the amplified result through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector and a reference vector; determining a value vector based on the amplification result; Processing the query vector, the key vector, and the value vector through a long short-term memory network of the pre-trained friction compensation model under the guidance of the amplification result to obtain a first processing result; Performing group normalization processing on the first processing result using a pre-trained friction compensation model to obtain a second normalized result; Performing feature fusion on the second normalization result, the reference vector, and the magnification result to obtain a feature fusion result; Reducing the feature fusion result according to a preset ratio to obtain a reduced result; Processing the reduced result through a Kolmogorov-Arnold layer to obtain a second processed result; Performing channel normalization processing on the second processing result using a pre-trained friction compensation model to obtain a predicted compensation friction force; The manipulator friction compensation model is obtained by adjusting the parameters of the pre-trained friction compensation model based on the difference between the standard compensation friction force and the predicted compensation friction force, and continuously training the pre-trained friction compensation model using the joint friction compensation data of the robot and the standard compensation friction force corresponding to the joint friction compensation data; The friction error of the robot is compensated by the robot friction compensation model.

2. The method according to claim 1, characterized in that The Kolmogorov-Arnold network includes a 1x4 convolution, a 1x1 convolution, a 4x1 convolution, a Kolmogorov-Arnold layer, a summation layer and an activation function, and the amplification result is processed through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector and a reference vector, including: The amplified result is respectively subjected to the 1x4 convolution, the 1x1 convolution and the 4x1 convolution to obtain a first convolution result, a second convolution result and a third convolution result; Summing the first convolution result, the second convolution result and the third convolution result through the summation layer to obtain a summation result; Inputting the summation result into the Kolmogorov-Arnold layer to obtain a nonlinear learning result; Fusing the nonlinear learning result and the amplification result to obtain a nonlinear fusion result; The nonlinear fusion result is obtained by using an activation function to obtain the query vector, the key vector and the reference vector.

3. The method according to claim 2, characterized in that The activation function is expressed by the following formula: LeakySigmoid(x)=x(Sigmoid(x)+αmax(0,-x)); Wherein, LeakySigmoid(x) represents an activation function, α represents a constant, and x represents the input of the activation function. .

4. The method according to claim 1, characterized in that: The parameters of the long short-term memory network are expressed by the following formula: ； Where t represents time, represents the memory state at time t, represents the first intermediate quantity at time t, represents the memory state at time t-1, represents the second intermediate quantity at time t, represents the value vector at time t, represents the key vector at time t; ； Among them, t represents time, n represents time series, represents the time series corresponding to time t, represents the first intermediate quantity at time t, represents the time series at time t-1, represents the second intermediate quantity at time t, represents the key vector at time t; ； Where t represents time, represents the measured value of the state quantity at time t, represents the state quantity at time t, represents the third intermediate quantity, Related to the query vector at time t.

5. The method according to claim 4, characterized in that The error value of the state quantity at time t is expressed by the following formula: ； Where t represents time, represents the measured value of the state quantity at time t, represents the memory state at time t, represents the query vector at time t, Represents the time series corresponding to time t.

6. The method according to claim 5, characterized in that The query vector, key vector and reference vector are expressed by the following formula: ； Where t represents time, represents the query vector at time t, represents the input value at time t, represents the first weight, represents the first component; ； Where t represents time, Indicates that the dimension of the query vector q is d, represents the key vector at time t, represents the input value at time t, represents the second weight, represents the second component; ； Where t represents time, represents the value vector at time t, represents the input value at time t, represents the third weight, Represents the third component.

7. The method according to claim 6, characterized in that The first intermediate amount, the second intermediate amount and the third intermediate amount are expressed by the following formula: ； Where t represents time, represents the first intermediate quantity at time t, represents the measured value of the first intermediate quantity at time t; ； in, represents the first parameter corresponding to the first intermediate quantity, represents the second parameter corresponding to the first intermediate quantity, represents the parameter corresponding to time t; ； Where t represents time, represents the second intermediate quantity at time t, represents the measured value of the second intermediate quantity at time t; ； in, represents the first parameter corresponding to the second intermediate quantity, represents the second parameter corresponding to the second intermediate quantity, represents the parameter corresponding to time t; ； Where t represents time, represents the third intermediate quantity at time t, represents the measured value of the third intermediate quantity at time t; ； in, represents the first parameter corresponding to the third intermediate quantity, represents the second parameter corresponding to the third intermediate quantity, Represents the input value at time t.

8. A device for compensating friction error of a robot arm, characterized in that: include: An acquisition module, used for acquiring joint friction force compensation data of the robot and a standard compensation friction force corresponding to the joint friction force compensation data, wherein the joint friction force compensation data includes joint curvature, joint output power, joint mass and joint length; A feature extraction module, used for performing feature extraction on the joint friction force compensation data to obtain a feature extraction result; A first normalization module, used to perform layer normalization processing on the feature extraction result using a pre-trained friction compensation model to obtain a first normalized result; An upper projection module, used for magnifying the first normalized result by a preset multiple to obtain a magnified result; A nonlinear learning module, used for processing the amplification result through the Kolmogorov-Arnold network of the pre-trained friction compensation model to obtain a query vector, a key vector and a reference vector; A determination module, configured to determine a value vector based on the amplification result; A first processing module, configured to process the query vector, the key vector and the value vector through a long short-term memory network of the pre-trained friction compensation model under the guidance of the amplification result to obtain a first processing result; A second normalization module, used for performing group normalization processing on the first processing result using a pre-trained friction compensation model to obtain a second normalized result; A fusion module, used for performing feature fusion on the second normalization result, the reference vector, and the amplification result to obtain a feature fusion result; A lower projection module, used for reducing the feature fusion result according to a preset ratio to obtain a reduced result; A second processing module, used for processing the reduction result through a Kolmogorov-Arnold layer to obtain a second processing result; a third normalization module, configured to perform channel normalization processing on the second processing result using a pre-trained friction compensation model to obtain a predicted compensation friction force; A training module, configured to adjust the parameters of the pre-trained friction compensation model based on the difference between the standard compensation friction force and the predicted compensation friction force, and continuously train the pre-trained friction compensation model using the joint friction compensation data of the robot and the standard compensation friction force corresponding to the joint friction compensation data to obtain a manipulator friction compensation model; The compensation module is used to compensate the friction error of the robot arm through the robot arm friction compensation model.

9. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store the computer program; The processor is used to execute the computer program to implement the method for compensating the friction error of the robot arm as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, the method for compensating the friction error of a robot arm as described in any one of claims 1 to 7 is implemented.

Images