High-precision calibration method and device for multi-degree-of-freedom serial manipulator

Abstract

The present invention discloses a high-precision calibration method and device for a multi-degree-of-freedom serial robot, which relates to the field of mechanical control technology, and the main purpose is to solve the problem of poor accuracy of the existing multi-degree-of-freedom serial robot motion position calibration. It includes: obtaining the angle data of the motion joint angle of the multi-degree-of-freedom serial robot and the nominal end position of the multi-degree-of-freedom serial robot; predicting the angle data and the nominal end position based on a convolutional neural network that has completed model training to obtain a compensated end position, and the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on the actual position sample of the multi-degree-of-freedom serial robot; when the evaluation value of the compensated end position matches the preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot is completed.

Description

Technical Field

[0001] The present invention relates to the field of mechanical control technology, and in particular to a high-precision calibration method and device for a multi-degree-of-freedom serial robot arm. Background Art

[0002] With the advancement of my country's intelligent manufacturing industry model, robot technology with the ability of offline programming and online teaching operation is very important for the intelligent and digital process of manufacturing. The research and development technology of multi-degree-of-freedom serial robot arms, represented by industrial robots, has great strategic significance for improving industrial equipment capabilities and promoting the transformation and upgrading of new production methods. This puts forward new requirements for the absolute position accuracy of the end of the multi-degree-of-freedom serial robot arm. Among them, the absolute position accuracy of the end of the industrial robot can be effectively improved through kinematic level calibration, so as to use error correction compensation for the robot.

[0003] At present, when using the kinematic calibration method to calibrate the position of a robot as a multi-degree-of-freedom serial manipulator, the least squares method is usually used to derive the kinematic parameters, that is, the least squares method assumes that the robot's kinematic model is geometrically accurate, and then determines the robot's generalized inverse matrix, so that the robot's motion can be completely described by geometric parameters (such as joint length, distance between joints, etc.), using model-based kinematic calibration or model-free kinematic calibration. However, when using the least squares method to solve the kinematic parameters, the robot's motion is also affected by many non-geometric factors, such as joint transmission errors, flexibility, friction, etc. These factors will introduce additional non-geometric errors, and also make the least squares method unable to integrate multiple error factors, and unable to meet the needs of accurate calibration of multi-degree-of-freedom serial manipulators, thereby affecting the precise control of multi-degree-of-freedom serial manipulators. Summary of the invention

[0004] In view of this, the present invention provides a high-precision calibration method and device for a multi-degree-of-freedom serial robot arm, the main purpose of which is to solve the problem of poor accuracy in the motion position calibration of the existing multi-degree-of-freedom serial robot arm.

[0005] According to one aspect of the present invention, a high-precision calibration method for a multi-degree-of-freedom serial robot arm is provided, comprising:

[0006] Acquire angle data of motion joint angles of a multi-degree-of-freedom serial robot arm and a nominal end position of the multi-degree-of-freedom serial robot arm, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming;

[0007] Based on a convolutional neural network that has completed model training, position prediction is performed on the angle data and the nominal end position to obtain a compensated end position, wherein the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on actual position samples of a multi-degree-of-freedom serial manipulator;

[0008] When the evaluation value of the compensation end position matches a preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot arm is completed.

[0009] Furthermore, before obtaining the angle data of the motion joint angles of the multi-degree-of-freedom serial robot arm and the nominal end position of the multi-degree-of-freedom serial robot arm, the method further includes:

[0010] Constructing a kinematic model, and determining the end pose matrix of the multi-degree-of-freedom serial manipulator in the nominal kinematic model based on the coordinate transformation matrix;

[0011] Solving the attitude rotation matrix and the nominal end position from the end pose matrix;

[0012] A global error function is constructed based on the nominal end position and the actual end position, and the global error function is used as an optimization target to optimize the kinematic parameter compensation value of the nominal kinematic model by using sequential quadratic programming;

[0013] The kinematic parameter compensation values ​​obtained by solving the problem are used to compensate the nominal kinematic model, thereby completing the identification of the nominal kinematic model.

[0014] Further, constructing a global error function based on the nominal end position and the actual end position includes:

[0015] The end position error for compensation is defined by defining parameters based on the difference between the nominal end position and the actual end position calculated by the kinematic model;

[0016] Construct a global error function based on the end position error R ;

[0017] in, . for n The end position error of the multi-degree-of-freedom serial manipulator of the group calibration point, is the end position error vector r The transpose of and Respectively expressed as j The nominal end position and the actual end position of the robot arm are calculated using the identified kinematic model under the group calibration points.

[0018] Furthermore, the end position errorr It is expressed as:

[0019] , and They are represented by the nominal end position calculated using the identified kinematic model and the actual end position of the multi-DOF serial robot, respectively. , , They are the nominal joint angle, joint angle error compensation value and actual joint angle error of the multi-degree-of-freedom serial robot arm, , , are the nominal geometric parameters, geometric parameter error compensation values ​​and actual geometric parameter errors of the multi-DOF serial manipulator, respectively. The geometric parameters include joint torsion angle, link length and link offset. Represents the forward kinematics function of the robot arm.

[0020] Furthermore, taking the global error function as the optimization target and using sequential quadratic programming to optimize the kinematic parameter compensation value of the nominal kinematic model includes:

[0021] Taking the minimization of the global error function as the optimization goal, sequential quadratic programming is used to iteratively search for kinematic parameter compensation values ​​of the nominal kinematic model, wherein the kinematic parameter compensation values ​​include optimal estimates of rod length errors and joint angle errors.

[0022] Furthermore, before the convolutional neural network based on the completed model training performs position prediction on the angle data and the nominal end position to obtain the compensated end position, the method further includes:

[0023] Acquire actual position samples of the multi-degree-of-freedom serial manipulator, wherein the actual position samples include actual end position samples corresponding to different angle samples and different nominal end position samples;

[0024] Create a convolutional neural network including three convolutional layers and two fully connected layers, wherein each convolutional layer in the convolutional neural network is connected to an activation function and a normalization function;

[0025] The convolutional neural network is trained based on the actual position samples.

[0026] Furthermore, the method further comprises:

[0027] Comparing the compensated end position with the acquired actual end position to obtain an evaluation value;

[0028] When the evaluation value does not match a preset calibration threshold, the network structure and hyperparameters in the convolutional neural network are adjusted to recalibrate.

[0029] According to another aspect of the present invention, a high-precision calibration device for a multi-degree-of-freedom serial robot arm is provided, comprising:

[0030] An acquisition module is used to acquire angle data of motion joint angles of a multi-degree-of-freedom serial robot arm and a nominal end position of the multi-degree-of-freedom serial robot arm, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming;

[0031] A prediction module, used for predicting the angle data and the nominal end position based on a convolutional neural network that has completed model training, so as to obtain a compensated end position, wherein the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on actual position samples of a multi-degree-of-freedom serial manipulator;

[0032] The calibration module is used to complete the motion position calibration of the multi-degree-of-freedom serial robot arm when the evaluation value of the compensation end position matches a preset calibration threshold.

[0033] Furthermore, the device also includes:

[0034] A first construction module is used to construct a kinematic model and determine the end pose matrix of the multi-degree-of-freedom serial manipulator in the nominal kinematic model based on a coordinate transformation matrix;

[0035] A solution module, used for solving the attitude rotation matrix and the nominal end position from the end attitude matrix;

[0036] A second construction module is used to construct a global error function based on the nominal end position and the actual end position, and use the global error function as an optimization target to optimize the kinematic parameter compensation value of the nominal kinematic model by using sequential quadratic programming;

[0037] The compensation module is used to compensate the nominal kinematic model with the kinematic parameter compensation value obtained by solving the problem, so as to complete the identification of the nominal kinematic model.

[0038] Furthermore, the second construction module is specifically used to define the end position error of the compensation by defining the parameters based on the difference between the nominal end position and the actual end position calculated by the kinematic model; and construct a global error function based on the end position error. R ;in, , for nThe end position error of the multi-degree-of-freedom serial manipulator of the group calibration point, is the end position error vector r The transpose of and Respectively expressed as j The nominal end position and the actual end position of the robot arm are calculated using the identified kinematic model under the group calibration points.

[0039] Furthermore, the end position error r It is expressed as:

[0040] , and They are represented by the nominal end position calculated using the identified kinematic model and the actual end position of the multi-DOF serial robot, respectively. , , They are the nominal joint angle, joint angle error compensation value and actual joint angle error of the multi-degree-of-freedom serial robot arm, , , are the nominal geometric parameters, geometric parameter error compensation values ​​and actual geometric parameter errors of the multi-DOF serial manipulator, respectively. The geometric parameters include joint torsion angle, link length and link offset. Represents the forward kinematics function of the robot arm.

[0041] Furthermore, the solution module is specifically used to take the minimization of the global error function as the optimization goal, and use sequential quadratic programming to iteratively search for the kinematic parameter compensation value of the nominal kinematic model, wherein the kinematic parameter compensation value includes the optimal estimate of the rod length error and the joint angle error.

[0042] Furthermore, the device further comprises: a third creation module, a training module,

[0043] The acquisition module is further used to acquire actual position samples of the multi-degree-of-freedom serial manipulator, wherein the actual position samples include actual end position samples corresponding to different angle samples and different nominal end position samples;

[0044] The third creation module is further used to create a convolutional neural network comprising three convolutional layers and two fully connected layers, wherein each convolutional layer in the convolutional neural network is connected to an activation function and a normalization function;

[0045] The training module is used to perform model training on the convolutional neural network based on the actual position samples.

[0046] Furthermore, the device also includes:

[0047] A comparison module, used for comparing the compensated end position with the acquired actual end position to obtain an evaluation value;

[0048] An adjustment module is used to adjust the network structure and hyperparameters in the convolutional neural network to recalibrate when the evaluation value does not match a preset calibration threshold.

[0049] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned high-precision calibration method for a multi-degree-of-freedom serial robot arm.

[0050] According to another aspect of the present invention, there is provided a terminal, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0051] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned high-precision calibration method for a multi-degree-of-freedom serial robot arm.

[0052] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0053] The present invention provides a high-precision calibration method and device for a multi-degree-of-freedom serial robot. Compared with the prior art, the embodiment of the present invention obtains the angle data of the motion joint angle of the multi-degree-of-freedom serial robot and the nominal end position of the multi-degree-of-freedom serial robot, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming; a convolutional neural network based on the completed model training performs position prediction on the angle data and the nominal end position to obtain a compensated end position. The convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on the actual position samples of the multi-degree-of-freedom serial robot; when the evaluation value of the compensation end position matches the preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot is completed, which can fully compensate for the non-geometric parameter errors of the robot and avoid the parameter solution differences caused by various error factors, thereby achieving the purpose of combining the kinematic model with the machine learning algorithm to calibrate the position of the multi-degree-of-freedom serial robot, thereby improving the accuracy of the position calibration of the multi-degree-of-freedom serial robot and improving the precise control of the multi-degree-of-freedom serial robot.

[0054] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0056] Figure 1 A flow chart of a high-precision calibration method for a multi-degree-of-freedom serial robot arm provided by an embodiment of the present invention is shown;

[0057] Figure 2 A flow chart of a high-precision kinematic calibration method for a robotic arm integrating kinematic parameter identification and convolutional neural network provided by an embodiment of the present invention is shown;

[0058] Figure 3 A schematic diagram of constructing a kinematic model of a robotic arm provided by an embodiment of the present invention is shown;

[0059] Figure 4 A schematic diagram of a convolutional neural network structure provided by an embodiment of the present invention is shown;

[0060] Figure 5 A block diagram of a high-precision calibration device for a multi-degree-of-freedom serial robot arm provided by an embodiment of the present invention is shown;

[0061] Figure 6 A schematic diagram of the structure of a terminal provided by an embodiment of the present invention is shown.

[0062] Description of reference numerals:

[0063] 1-Joint i-1; 2-Link i-1; 3-Key i; 4-Link i. DETAILED DESCRIPTION

[0064] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0065] The embodiment of the present invention provides a high-precision calibration method for a multi-degree-of-freedom serial robot arm, such as Figure 1 As shown, the method includes:

[0066] 101. Obtain angle data of motion joint angles of a multi-degree-of-freedom serial robot arm and a nominal end position of the multi-degree-of-freedom serial robot arm.

[0067] In the embodiment of the present invention, the current execution end, as the execution subject for calibrating the position movement of the robot arm, can be a cloud server or a terminal device, and further, the movement of the robot arm is collected in real time for calibration. Among them, the robot arm can be a device that moves through a multi-joint connecting rod, and further, the current execution end obtains the angle data of the motion joint angle of the robot arm, as well as the nominal end position. Among them, the nominal end position is calculated based on the identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming. The nominal kinematic model is constructed based on the robot arm according to the kinematic principle, and the embodiment of the present invention does not make specific limitations. In addition, the angle data includes the sine value and cosine value of the motion joint angle.

[0068] 102. Perform position prediction on the angle data and the nominal end position based on the convolutional neural network that has completed model training to obtain a compensated end position.

[0069] In an embodiment of the present invention, the current execution end uses a convolutional neural network that has completed model training to predict the angle data and the nominal end position to obtain the compensated end position. At this time, the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on the actual position samples of the robotic arm, thereby using this convolutional neural network for prediction.

[0070] It should be noted that if Figure 2 In the calibration process shown in the figure, after the convolutional neural network is trained, the nominal end position obtained by the identified kinematic model is P i And angle data (cos Q , sin Q ) as model input, Q is the joint angle. After position prediction, the compensated end position is obtained. P c .

[0071] 103. When the evaluation value of the compensation end position matches a preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot arm is completed.

[0072] In the embodiment of the present invention, after the current execution end obtains the compensation end position, the compensation end position is evaluated to obtain an evaluation value. If the evaluation value matches the preset calibration threshold, it means that the compensation end position has an accurate supplementary effect, and thus the motion position calibration of the robot arm is completed. The preset calibration threshold can be configured based on the calibration requirements of the robot arm, and the embodiment of the present invention does not specifically limit it.

[0073] In another embodiment of the present invention, for further definition and explanation, before the step of obtaining the angle data of the motion joint angle of the multi-degree-of-freedom serial robot arm and the nominal end position of the multi-degree-of-freedom serial robot arm, the method further includes:

[0074] Constructing a kinematic model, and determining the end pose matrix of the multi-degree-of-freedom serial manipulator in the nominal kinematic model based on the coordinate transformation matrix;

[0075] Solving the attitude rotation matrix and the nominal end position from the end pose matrix;

[0076] A global error function is constructed based on the nominal end position and the actual end position, and the global error function is used as an optimization target to optimize the kinematic parameter compensation value of the nominal kinematic model by using sequential quadratic programming;

[0077] The kinematic parameter compensation values ​​obtained by solving the problem are used to compensate the nominal kinematic model, thereby completing the identification of the nominal kinematic model.

[0078] In order to combine kinematics with machine learning models to achieve the purpose of accurate robot arm calibration, the current execution end pre-constructs a kinematic model. Among them, the kinematic model is a mathematical model constructed based on the joint node position of the robot arm as the research object, such as the kinematic model of the robot arm is constructed by the Denavit-Hartenberg modeling method (abbreviated as: DH modeling method) as a nominal kinematic model. Specifically, the actual position of the end of the robot arm is measured under several sets of robot arm configurations. At the same time, the nominal end position of the robot arm under the corresponding configuration is calculated according to the nominal kinematic DH model of the robot arm. The nominal kinematic DH model describes the relative motion relationship between each rod and the end of the robot arm, and establishes a corresponding coordinate system for each rod in the robot arm motion chain. When the kinematic model is constructed, the current execution end will determine the end pose matrix of the robot arm in the nominal kinematic model based on the coordinate transformation matrix corresponding to the kinematic model, and the geometric parameters of the robot arm can be calibrated. The geometric parameter error of the robot arm is caused by the error between the nominal kinematic model DH and the actual parameters, including the rod length error. , torsion angle error , offset error and angular error ,in, i Indicates i The joint coordinate system, i =1,2,…, n , n is the degree of freedom of the robot arm, such as Figure 3 Schematic diagram of the DH model of the robot arm configuration shown. In a specific implementation scenario, the calculation of the nominal end position of the robot arm is based on kinematics. Therefore, the coordinate transformation matrix is ​​used to obtain the coordinate system of the adjacent joints of the robot arm. arrive The conversion relationship is:

[0079] ;

[0080] in, are DH parameters, representing the i The link length, torsion angle, link offset and joint rotation in the joint coordinate system are Represents the adjacent joint coordinate system arrive The transformation matrix, is the rotation matrix representing the rotation around z Axis rotation angle q i , is the translation matrix along z Axis moving distance d i Therefore, the expression of the robot end effector in the world coordinate system is:

[0081] ;

[0082] in, is the posture rotation matrix of the robot end effector in the world coordinate system, This is the nominal end position of the desired robotic arm.

[0083] It should be noted that in order to find the optimal parameters in the motion model, a global error function is constructed using the actual measured values ​​of the end position of the robot and its nominal value, and the kinematic calibration problem is transformed into a parameter compensation optimization problem. In this process, sequential quadratic programming is used to optimize the kinematic parameter compensation values ​​of the nominal kinematic model, and the kinematic parameter compensation values ​​obtained are used to compensate the nominal kinematic model to complete the identification of the nominal kinematic model. In addition, the actual end position can be measured using the API OMNITRAC2 laser tracker to measure the coordinates of the reflective target ball installed at the end of the robot to obtain the actual end positions of several groups of robot arms with different configurations.

[0084] In another embodiment of the present invention, for further definition and explanation, the step of constructing a global error function based on the nominal end position and the actual end position includes:

[0085] The end position error for compensation is defined by defining parameters based on the difference between the nominal end position and the actual end position calculated by the kinematic model;

[0086] Construct a global error function based on the end position error R .

[0087] In order to improve the efficiency of solving the nominal kinematic model, when the current execution end constructs the global error function, in order to compensate the nominal kinematic model with the kinematic parameter compensation value obtained by the solution and complete the identification of the nominal kinematic model, when the current execution end constructs the global error function, specifically, the end position error of the compensation is defined by the parameter identification through the difference between the nominal end position calculated by the kinematic model and the actual end position, so as to construct the global error function based on the end position error. R .

[0088] Specifically, the end position error of the manipulator compensated by parameter identification is defined as the difference between the nominal end position calculated by the identified kinematic model and the actual end position of the manipulator, where: , for n The end position error of the multi-degree-of-freedom serial manipulator of the group calibration point, is the end position error vector r The transpose of and Respectively expressed as j The nominal end position and the actual end position of the robot arm are calculated using the identified kinematic model under the group calibration points.

[0089] Furthermore, the end position error r It is expressed as:

[0090] , and They are represented by the nominal end position calculated using the identified kinematic model and the actual end position of the multi-DOF serial robot, respectively. , , They are the nominal joint angle, joint angle error compensation value and actual joint angle error of the multi-degree-of-freedom serial robot arm, , , are the nominal geometric parameters, geometric parameter error compensation values ​​and actual geometric parameter errors of the multi-DOF serial manipulator, respectively. The geometric parameters include joint torsion angle, link length and link offset. In another embodiment of the present invention, for further definition and explanation, the step of taking the global error function as the optimization target and using sequential quadratic programming to optimize the kinematic parameter compensation value of the nominal kinematic model includes:

[0091] Taking the minimization of the global error function as the optimization goal, sequential quadratic programming is used to iteratively search for kinematic parameter compensation values ​​of the nominal kinematic model, wherein the kinematic parameter compensation values ​​include optimal estimates of rod length errors and joint angle errors.

[0092] In order to seek the optimal compensation parameters, thereby improving the accuracy of identifying the kinematic model and improving the accuracy of the robot position calibration, the current execution end, when seeking to solve the kinematic parameter compensation value, specifically, takes the minimization of the global error function as the optimization goal, and uses the Sequential Quadratic Programming (SQP) algorithm to optimize the parameter error compensation value of the nominal kinematic model DH. At this time, the geometric error compensation target is determined as the minimization of the global error function. After the geometric parameter calibration process, the identified kinematic model can be obtained, so that after the parameter compensation of the robot arm, a convolutional neural network is established to compensate for the parameter error of the robot arm.

[0093] It should be noted that when optimizing the parameter compensation value of the nominal kinematic model based on the SQP algorithm, the optimization objective is the global error function Minimize and iteratively search for the optimal estimate of the rod length error and joint angle error. The formula is as follows:

[0094] ;

[0095] Among them, min is the minimum operator, , Indicates the iteration point k The joint angle error and parameter error of the robot arm at , the parameter error is specifically the rod length error; represents the gradient operator, and are used to limit the range of the identified joint angle error and rod length error. i The constraints of the equation and j Inequality constraints, and denote the condition sets of equation constraints and inequality constraints respectively, It represents the Hessian matrix at the iteration point, so that the joint angle error and rod length error obtained in the parameter calibration process are compensated to the nominal kinematic model DH to obtain the identified nominal kinematic model.

[0096] In another embodiment of the present invention, for further definition and explanation, before the step of performing position prediction on the angle data and the nominal end position based on the convolutional neural network that has completed model training to obtain the compensated end position, the method further includes:

[0097] Obtain actual position samples of multi-DOF serial manipulators;

[0098] Create a convolutional neural network with three convolutional layers and two fully connected layers;

[0099] The convolutional neural network is trained based on the actual position samples.

[0100] In order to effectively identify the position of the robot arm, a convolutional neural network containing three convolutional layers and two fully connected layers is pre-built by combining the kinematic model with the machine learning model, and the convolutional neural network model is trained in combination with the actual position samples of the robot arm. Among them, each convolutional layer in the convolutional neural network is connected to an activation function and a normalization function. The activation function and the normalization function are not specifically limited. The actual position samples include actual end position samples corresponding to different angle samples and different nominal end position samples. In a specific implementation scenario, the actual position samples can be obtained by collecting the actual robot arm. Considering that the periodicity of the joint angle may cause overfitting of the network training, the input of the training set is the sine and cosine values ​​of the joint angle and the nominal end position calculated by the identified kinematic model, and the output of the training set is the actual end position of the robot arm. Furthermore, the input of the convolutional neural network is the sine and cosine values ​​of the joint angle and the nominal end position calculated using the identified kinematic model. The network output is the end position of the robot after integrating the geometric and non-geometric parameter error compensation, so as to compensate for the non-geometric parameter errors such as gap, temperature, thermal expansion, and hysteresis effect, and eliminate the influence of non-geometric parameters on position calibration.

[0101] It should be noted that in the process of calibrating the position of the robot, the input of the convolutional neural network model is the sine and cosine values ​​of the joint angle and the nominal end position calculated by the identified kinematic model. The convolutional neural network model outputs the end position of the robot arm after network compensation. , the end position after network compensation The actual end position of the robot arm Compare to evaluate the calibration effect.

[0102] In another embodiment of the present invention, for further definition and description, the steps further include:

[0103] Comparing the compensated end position with the acquired actual end position to obtain an evaluation value;

[0104] When the evaluation value does not match a preset calibration threshold, the network structure and hyperparameters in the convolutional neural network are adjusted to recalibrate.

[0105] In order to improve the accuracy of convolutional neural network calibration of the robotic arm, during the evaluation process, the calibration effect can be improved by adjusting the neural network structure and hyperparameters. Specifically, the current execution end compares the compensated end position with the actual end position collected to obtain an evaluation value. At this time, the network structure of the convolutional neural network consists of 3 convolutional layers and 2 fully connected hidden layers. The size of the first convolution kernel is 32×1, and the size of the remaining kernels is 3×1. A ReLU activation function and a BN function are attached to each convolution layer to accelerate the training of the network and extract nonlinearity from the data. Figure 4 The schematic diagram of the convolutional neural network structure is shown in FIG. The parameters of the convolutional layer and the fully connected layer are shown in Table 1. The complete calibration process of the high-precision kinematic calibration method of the manipulator integrating kinematic parameter identification and convolutional neural network is shown in FIG. Figure 2 Furthermore, when the evaluation value does not match the preset calibration threshold, the network structure and hyperparameters in the convolutional neural network are adjusted to recalibrate. The preset calibration threshold can be based on the numerical content of the configuration threshold for different robotic arms, such as robotic arms of different models and different application scenarios, which is not specifically limited in the embodiments of the present invention.

[0106] Table 1 CNN model details

[0107]

[0108] In a specific application scenario of an embodiment of the present invention, after the current execution end completes the motion position calibration of the multi-degree-of-freedom serial robot arm, it calls up the operation control strategy of the robot arm operation object, and calculates each operation position point corresponding to the operation control strategy based on the calibrated motion position; if the movement distance of each operation position point is greater than the preset calibration error, the robot arm motion position calibration of steps 101-103 is re-executed; if the movement distance of each operation position point is less than or equal to the preset calibration error, the operation of driving the robot arm according to the operation control strategy is started.

[0109] In the embodiment of the present invention, after the motion position calibration of the robot arm is completed, in order to accurately and effectively use the calibrated robot arm for production, thereby realizing the precise use of the calibrated robot arm, the current execution end first retrieves the operation control strategy of the robot arm operation object after completing the motion position calibration. At this time, the operation control strategy is used to characterize the control method used in the industrial production process of the robot arm, including but not limited to PID control in automatic control, control flow compilation using a programmable logic controller PLC, logic control, fuzzy control, etc., which can be pre-configured according to different industrial needs, so as to predict the control of the robot arm according to the operation control strategy, and the embodiment of the present invention does not make specific restrictions. Among them, the operation position points corresponding to the operation control strategy are calculated to use the compensation end position and the nominal end position used by the robot arm during calibration as the initial position for driving the robot arm, and the operation control strategy is embedded in the simulation software, such as the simulation of the robot arm in the MATLAB software according to the pre-configured operation control strategy to simulate and simulate, and obtain multiple operation position points during the operation of the robot arm. Among them, when obtaining each operation position point, the movement distance of the robot arm during the movement process can be used as a basis for judging whether the calibration of the robot arm is accurate through the rangefinder strategy. Specifically, if the movement distance of each operating position point is greater than the preset calibration error, it means that there is an error in the calibration of the robot arm, and the robot arm movement position calibration of steps 101-103 is re-executed. If the movement distance of each operating position point is less than or equal to the preset calibration error, the operation of driving the robot arm according to the operation control strategy is started, wherein the preset calibration error can be configured based on the length of the robot arm, the control method executed by the robot arm, and the accuracy requirement of the robot arm's operation object, and the embodiment of the present invention does not make specific limitations.

[0110] The embodiment of the present invention provides a high-precision calibration method for a multi-degree-of-freedom serial robot. Compared with the prior art, the embodiment of the present invention obtains the angle data of the motion joint angles of the multi-degree-of-freedom serial robot and the nominal end position of the multi-degree-of-freedom serial robot, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming; a convolutional neural network based on the completed model training performs position prediction on the angle data and the nominal end position to obtain a compensated end position. The convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on the actual position samples of the multi-degree-of-freedom serial robot; when the evaluation value of the compensation end position matches the preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot is completed, which can fully compensate for the non-geometric parameter errors of the robot and avoid the parameter solution differences caused by various error factors, thereby achieving the purpose of combining the kinematic model with the machine learning algorithm to calibrate the position of the multi-degree-of-freedom serial robot, thereby improving the accuracy of the position calibration of the multi-degree-of-freedom serial robot and improving the precise control of the multi-degree-of-freedom serial robot.

[0111] Furthermore, as a response to the above Figure 1 The embodiment of the present invention provides a high-precision calibration device for a multi-degree-of-freedom serial robot arm, such as Figure 5 As shown, the device comprises:

[0112] An acquisition module 21 is used to acquire angle data of motion joint angles of a multi-degree-of-freedom serial robot arm and a nominal end position of the multi-degree-of-freedom serial robot arm, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming;

[0113] A prediction module 22, configured to perform position prediction on the angle data and the nominal end position based on a convolutional neural network that has completed model training, to obtain a compensated end position, wherein the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on actual position samples of a multi-degree-of-freedom serial manipulator;

[0114] The calibration module 23 is used to complete the motion position calibration of the multi-degree-of-freedom serial robot arm when the evaluation value of the compensation end position matches a preset calibration threshold.

[0115] Furthermore, the device also includes:

[0116] A first construction module is used to construct a kinematic model and determine the end pose matrix of the multi-degree-of-freedom serial manipulator in the nominal kinematic model based on a coordinate transformation matrix;

[0117] A solution module, used for solving the attitude rotation matrix and the nominal end position from the end attitude matrix;

[0118] A second construction module is used to construct a global error function based on the nominal end position and the actual end position, and use the global error function as an optimization target to optimize the kinematic parameter compensation value of the nominal kinematic model by using sequential quadratic programming;

[0119] The compensation module is used to compensate the nominal kinematic model with the kinematic parameter compensation value obtained by solving the problem, so as to complete the identification of the nominal kinematic model.

[0120] Furthermore, the second construction module is specifically used to define the end position error of the compensation by defining the parameters based on the difference between the nominal end position and the actual end position calculated by the kinematic model; and construct a global error function based on the end position error. R ;in, , for n The end position error of the multi-degree-of-freedom serial manipulator of the group calibration point, is the end position error vector r The transpose of and Respectively expressed as j The nominal end position and the actual end position of the robot arm are calculated using the identified kinematic model under the group calibration points.

[0121] Furthermore, the end position error r It is expressed as:

[0122] , and They are represented by the nominal end position calculated using the identified kinematic model and the actual end position of the multi-DOF serial robot, respectively. , , They are the nominal joint angle, joint angle error compensation value and actual joint angle error of the multi-degree-of-freedom serial robot arm, , , are the nominal geometric parameters, geometric parameter error compensation values ​​and actual geometric parameter errors of the multi-DOF serial manipulator, respectively. The geometric parameters include joint torsion angle, link length and link offset. Represents the forward kinematics function of the manipulator. Further, the solution module is specifically used to take the minimization of the global error function as the optimization goal, and use sequential quadratic programming to iteratively search for the kinematic parameter compensation value of the nominal kinematic model, wherein the kinematic parameter compensation value includes the optimal estimation of the rod length error and the joint angle error.

[0123] Furthermore, the device further comprises: a third creation module, a training module,

[0124] The acquisition module is further used to acquire actual position samples of the multi-degree-of-freedom serial manipulator, wherein the actual position samples include actual end position samples corresponding to different angle samples and different nominal end position samples;

[0125] The third creation module is further used to create a convolutional neural network comprising three convolutional layers and two fully connected layers, wherein each convolutional layer in the convolutional neural network is connected to an activation function and a normalization function;

[0126] The training module is used to perform model training on the convolutional neural network based on the actual position samples.

[0127] Furthermore, the device also includes:

[0128] A comparison module, used for comparing the compensated end position with the acquired actual end position to obtain an evaluation value;

[0129] An adjustment module is used to adjust the network structure and hyperparameters in the convolutional neural network to recalibrate when the evaluation value does not match a preset calibration threshold.

[0130] The embodiment of the present invention provides a high-precision calibration device for a multi-degree-of-freedom serial robot. Compared with the prior art, the embodiment of the present invention obtains the angle data of the motion joint angles of the multi-degree-of-freedom serial robot and the nominal end position of the multi-degree-of-freedom serial robot, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming; a convolutional neural network based on the completed model training performs position prediction on the angle data and the nominal end position to obtain a compensated end position. The convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on the actual position samples of the multi-degree-of-freedom serial robot; when the evaluation value of the compensation end position matches the preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot is completed, which can fully compensate for the non-geometric parameter errors of the robot and avoid the parameter solution differences caused by various error factors, thereby achieving the purpose of combining the kinematic model with the machine learning algorithm to calibrate the position of the multi-degree-of-freedom serial robot, thereby improving the accuracy of the position calibration of the multi-degree-of-freedom serial robot and improving the precise control of the multi-degree-of-freedom serial robot.

[0131] According to one embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the high-precision calibration method for a multi-degree-of-freedom serial robot arm in any of the above method embodiments.

[0132] Figure 6 A schematic diagram of the structure of a terminal provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the terminal.

[0133] like Figure 6 As shown, the terminal may include: a processor (processor) 302 , a communication interface (Communications Interface) 304 , a memory (memory) 306 , and a communication bus 308 .

[0134] The processor 302 , the communication interface 304 , and the memory 306 communicate with each other via the communication bus 308 .

[0135] The communication interface 304 is used to communicate with other devices such as clients or other servers.

[0136] The processor 302 is used to execute the program 310, and specifically can execute the relevant steps in the above-mentioned embodiment of the high-precision calibration method for a multi-degree-of-freedom serial robot arm.

[0137] Specifically, the program 310 may include program codes, which include computer operation instructions.

[0138] The processor 302 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiment of the present invention. The one or more processors included in the terminal may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0139] The memory 306 is used to store the program 310. The memory 306 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0140] The program 310 may be specifically configured to enable the processor 302 to perform the following operations:

[0141] Acquire angle data of motion joint angles of a multi-degree-of-freedom serial robot arm and a nominal end position of the multi-degree-of-freedom serial robot arm, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming;

[0142] Based on a convolutional neural network that has completed model training, position prediction is performed on the angle data and the nominal end position to obtain a compensated end position, wherein the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on actual position samples of a multi-degree-of-freedom serial manipulator;

[0143] When the evaluation value of the compensation end position matches a preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot arm is completed.

[0144] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0145] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A high-precision calibration method for a multi-degree-of-freedom serial robot arm, characterized in that: include: Acquire angle data of motion joint angles of a multi-degree-of-freedom serial robot arm and a nominal end position of the multi-degree-of-freedom serial robot arm, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming; Based on a convolutional neural network that has completed model training, position prediction is performed on the angle data and the nominal end position to obtain a compensated end position, wherein the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on actual position samples of a multi-degree-of-freedom serial manipulator; When the evaluation value of the compensation end position matches a preset calibration threshold, the motion position calibration of the multi-degree-of-freedom serial robot arm is completed; Before obtaining the angle data of the motion joint angles of the multi-degree-of-freedom serial robot arm and the nominal end position of the multi-degree-of-freedom serial robot arm, the method further comprises: Constructing a kinematic model, and determining the end pose matrix of the multi-degree-of-freedom serial manipulator in the nominal kinematic model based on the coordinate transformation matrix; Solving the attitude rotation matrix and the nominal end position from the end pose matrix; A global error function is constructed based on the nominal end position and the actual end position, and the global error function is used as an optimization target to optimize the kinematic parameter compensation value of the nominal kinematic model by using sequential quadratic programming; The kinematic parameter compensation values ​​obtained by solving the problem are used to compensate the nominal kinematic model, thereby completing the identification of the nominal kinematic model; The constructing of a global error function based on the nominal end position and the actual end position comprises: The end position error for compensation is defined by defining parameters based on the difference between the nominal end position and the actual end position calculated by the kinematic model; Construct a global error function based on the end position error R ; in, , for n The end position error of the multi-degree-of-freedom serial manipulator of the group calibration point, is the end position error vector r The transpose of and Respectively expressed as j The nominal end position and the actual end position of the robot arm are calculated using the identified kinematic model under the group calibration points.

2. The method according to claim 1, characterized in that The end position error r It is expressed as: , and They are represented by the nominal end position calculated using the identified kinematic model and the actual end position of the multi-DOF serial robot, respectively. , , They are the nominal joint angle, joint angle error compensation value and actual joint angle error of the multi-degree-of-freedom serial robot arm, , , are the nominal geometric parameters, geometric parameter error compensation values ​​and actual geometric parameter errors of the multi-DOF serial manipulator, respectively. The geometric parameters include joint torsion angle, link length and link offset. , They represent the forward kinematic functions of the robotic arm respectively.

3. The method according to claim 1, characterized in that The method of taking the global error function as the optimization target and using sequential quadratic programming to optimize the kinematic parameter compensation value of the nominal kinematic model comprises: Taking the minimization of the global error function as the optimization goal, sequential quadratic programming is used to iteratively search for kinematic parameter compensation values ​​of the nominal kinematic model, wherein the kinematic parameter compensation values ​​include optimal estimates of rod length errors and joint angle errors.

4. The method according to claim 1, characterized in that: Before the convolutional neural network based on the completed model training performs position prediction on the angle data and the nominal end position to obtain the compensated end position, the method further includes: Acquire actual position samples of the multi-degree-of-freedom serial manipulator, wherein the actual position samples include actual end position samples corresponding to different angle samples and different nominal end position samples; Create a convolutional neural network including three convolutional layers and two fully connected layers, wherein each convolutional layer in the convolutional neural network is connected to an activation function and a normalization function; The convolutional neural network is trained based on the actual position samples.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Comparing the compensated end position with the acquired actual end position to obtain an evaluation value; When the evaluation value does not match a preset calibration threshold, the network structure and hyperparameters in the convolutional neural network are adjusted to recalibrate.

6. A high-precision calibration device for a multi-degree-of-freedom serial robot arm, characterized in that: include: An acquisition module is used to acquire angle data of motion joint angles of a multi-degree-of-freedom serial robot arm and a nominal end position of the multi-degree-of-freedom serial robot arm, wherein the nominal end position is calculated based on an identified nominal kinematic model, and the identified nominal kinematic model is identified by performing error compensation optimization on the nominal kinematic model based on sequential quadratic programming; A prediction module, used for predicting the angle data and the nominal end position based on a convolutional neural network that has completed model training, so as to obtain a compensated end position, wherein the convolutional neural network is obtained by training a convolutional neural network including three convolutional layers based on actual position samples of a multi-degree-of-freedom serial manipulator; A calibration module, configured to complete the motion position calibration of the multi-degree-of-freedom serial robot arm when the evaluation value of the compensation end position matches a preset calibration threshold; The device also includes: A first construction module is used to construct a kinematic model and determine the end pose matrix of the multi-degree-of-freedom serial manipulator in the nominal kinematic model based on a coordinate transformation matrix; A solution module, used for solving the attitude rotation matrix and the nominal end position from the end attitude matrix; A second construction module is used to construct a global error function based on the nominal end position and the actual end position, and use the global error function as an optimization target to optimize the kinematic parameter compensation value of the nominal kinematic model by using sequential quadratic programming; A compensation module, used for compensating the nominal kinematics model with the kinematics parameter compensation value obtained by solving the problem, so as to complete the identification of the nominal kinematics model; The second construction module is specifically used to define the end position error of the compensation by defining parameters based on the difference between the nominal end position and the actual end position calculated by the kinematic model; and to construct a global error function based on the end position error. R ;in, , for n The end position error of the multi-degree-of-freedom serial manipulator of the group calibration point, is the end position error vector r The transpose of and Respectively expressed as j The nominal end position and the actual end position of the robot arm are calculated using the identified kinematic model under the group calibration points.

7. A storage medium, characterized in that: The storage medium stores at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the high-precision calibration method for a multi-degree-of-freedom serial robot arm as described in any one of claims 1-5.

8. A terminal, characterized in that: include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the high-precision calibration method for a multi-degree-of-freedom serial robot arm as described in any one of claims 1-5.

Images