An industrial robot teaching method based on six-dimensional force sensor information learning

Abstract

The present invention discloses an industrial robot teaching method based on six-dimensional force sensing information learning, which belongs to the field of intelligent manufacturing. The present invention uses a machine learning method to train a neural network with the force conditions of the end during the teaching process of a six-axis collaborative robot as input and the end speed as output, and applies the trained neural network to the industrial robot teaching process, so that the industrial robot end has the motion characteristics of the collaborative robot end. The present invention can realize teaching by dragging the industrial robot end, which improves the problems of poor human-computer interaction and cumbersome teaching process during the industrial robot teaching process.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent manufacturing, and more specifically, is an industrial robot teaching method based on six-dimensional force sensing information learning. Background Art

[0002] With the development of science and technology, industrial robots play an increasingly important role in production and life. Under the general trend of automation and intelligence in manufacturing, higher requirements are placed on the performance of human-machine interaction of industrial robots. However, as an indispensable part of the production process, industrial robot teaching still has the problems of cumbersome teaching process and poor human-machine interaction.

[0003] Industrial robot teaching refers to the process of guiding the robot's end effector or operating a mechanical simulation device to enable the robot to complete the expected motion sequence, speed, posture and other actions and store the teaching content in the controller. In actual production, industrial robots often use remote control teaching, offline teaching and indirect teaching methods due to the particularity of their motor and reducer solutions; the first two methods require the use of a teaching pendant and a host computer program to implement them respectively, with poor interactivity and cumbersome teaching process, which is not conducive to improving production efficiency; and indirect teaching is often based on visual or force sensors for teaching. Considering the complexity of the working environment of industrial robots, teaching based on visual sensors cannot achieve good results under some complex working conditions; an industrial robot teaching method based on six-dimensional force sensing information learning is proposed to improve the problems of poor human-machine interactivity and cumbersome teaching process in the industrial robot teaching process. Summary of the invention

[0004] In order to simplify the teaching process of industrial robots and improve the problem of poor interactive experience during the teaching process of industrial robots, the present invention aims to provide an industrial robot teaching method based on six-dimensional force sensing information learning.

[0005] To achieve the above object, the technical solution adopted by the present invention mainly includes the following steps:

[0006] Step 1: Equip two robot devices, namely a six-axis industrial robot and a six-axis collaborative robot. The ends of the six-axis industrial robot and the six-axis collaborative robot are both connected to a six-dimensional force sensor;

[0007] Step 2: Establish a neural network model, and obtain the robot terminal motion characteristic library through multiple teaching of the six-axis collaborative robot, and use the motion characteristic library to train the established neural network model. The specific steps are as follows:

[0008] Step 2a, setting the six-axis collaborative robot teaching sampling period T1, and the operator drags the end of the six-axis collaborative robot for teaching. During the teaching process, the joint angle signal of the six-axis collaborative robot and the six-dimensional force information of the end are sampled once in each sampling period until a teaching task is completed;

[0009] The six-axis collaborative robot joint angle signal and the six-dimensional force information of the end are sampled, wherein the six-axis collaborative robot joint angle signal is the six joint angle data of the six-axis collaborative robot, and the six-dimensional force information is the contact force signal of the robot end along the three coordinate axes of x, y, and z and the torque signal around the three coordinate axes of x, y, and z;

[0010] Step 2b, repeating step 2a to complete multiple teaching tasks, and obtaining joint angle data and end force sensor data corresponding to multiple teaching tasks of the six-axis collaborative robot;

[0011] Step 2c: pre-process all the obtained force sensor data using Bezier curves to obtain smooth force data and force increment data of each sampling period; and calculate the end position of the six-axis collaborative robot and its end velocity of the corresponding sampling period through forward kinematics;

[0012] As a preference, all the obtained force sensor data are preprocessed using a Bezier curve. Specifically, the force sensor data is substituted into the Bezier curve formula for n times, where n is the quotient of the total teaching time S and the sampling period T1, and the Bezier curve is obtained. Then, the smoothed force data of the i-th sample point after processing is Where i is an integer in the range [0, S / T1];

[0013] As a preferred method, the force increment data calculation method of each sampling period is:

[0014] As a preferred embodiment, the end position of the six-axis collaborative robot is calculated by forward kinematics and described by x, y, z coordinates and zyx Euler angles;

[0015] Step 2d, establishing a robot terminal motion characteristic library, which is composed of the smooth force data obtained in step 2c, the force increment data of each sampling period, and the terminal speed;

[0016] Step 2e, constructing a neural network with the force sensor data and each cycle force increment data as input and the terminal velocity as output; and using the data in the motion characteristic library established in step 2d to train the neural network to obtain the final regression model;

[0017] Preferably, the neural network is a BP neural network, which consists of an input layer, two hidden layers and an output layer: the number of nodes in the input layer is 12, and the corresponding inputs are the contact force signals of the robot end along the three coordinate axes of x, y, and z, the torque signals around the three coordinate axes of x, y, and z, and the increments of each force signal and torque signal relative to the previous sampling period; the number of nodes in the two middle layers is 10; the number of nodes in the output layer is 6, corresponding to the speed of the robot end along the three coordinate axes of x, y, and z and the angular velocity around the three coordinate axes of x, y, and z; the Relu function is selected as the activation function, and the mean square error function is selected as the loss function;

[0018] Step 3: Apply the final regression model obtained through training to the teaching process of the six-axis industrial robot. The specific steps are as follows:

[0019] Step 3a, setting the six-axis industrial robot teaching sampling period T2, and having the operator drag the end of the six-axis industrial robot, while the six-dimensional force sensor connected to the end collects the six-dimensional force information of the end in real time;

[0020] Step 3b: In each sampling period T2, the collected terminal six-dimensional force information f e Processing and generating the end driving force F and the force increment of each sampling period, and inputting them into the final regression model in real time, and outputting the expected speed of the end of the six-axis industrial robot corresponding to each sampling period T2;

[0021] As a preferred method, the collected terminal six-dimensional force information f e Processing and generating the final driving force F d and the force increment of each sampling period. The specific method is to use a piecewise function H(f e ) for the terminal driving force F d Mapping, that is, F d =H(f e ), where H(f e ) is expressed as follows:

[0022]

[0023] Among them, K P is the gain coefficient, and its value range is (0, +∞); K t is the motion threshold, and its value is related to the maximum value of the smoothing force in step 2c.

[0024] The six-dimensional force information f e The purpose of the processing is to prevent excessive external force input from causing the instantaneous speed of the end of the six-axis industrial robot to be too large;

[0025] Step 3c, using the obtained desired velocity of the end of the six-axis industrial robot, solving the desired posture of the end of the six-axis industrial robot, and finally solving it through inverse kinematics to complete a stepping motion;

[0026] As a preferred method, the desired posture of the end of the six-axis industrial robot is solved, specifically, the desired posture at time k is calculated. It is represented as the terminal pose P estimated at time k-1 k-1 The change of the end position of the six-axis industrial robot at time k is ΔP k The sum of the two, as shown in formula (3); and the estimated terminal pose P at time k-1 k-1 , then use the expected posture at time k-1 and the measured pose Z at time k-1 k-1 Characterized as formula (1), where k w is the confidence coefficient, and its value range is [0, 1]. The closer the confidence is to 0, the more reliable the expected posture is, and the closer the confidence is to 1, the more reliable the measured posture is. The change of the terminal posture of the six-axis industrial robot at time k, ΔP k It can be solved by formula (2), where v k With v k-1 are the terminal velocities of the six-axis industrial robot output by the final regression model at time k and time k-1, respectively, and the sampling period T2 is a known quantity; the mathematical model of the above process is as follows:

[0027]

[0028] ΔP k =(v k +v k-1 )·T2 / 2 (2)

[0029]

[0030] In the actual teaching process, first obtain the expected posture at time k-1 and the measured pose Z k-1 , calculate the estimated terminal pose P at time k-1 k-1 ; Then the six-dimensional force sensor signal at the end of the six-axis industrial robot collected at time k is processed to obtain the end driving force F d Then input it into the final regression model to obtain the terminal velocity v of the six-axis industrial robot k , and calculate the change in the end position of the six-axis industrial robot at time k ΔP k ; The estimated terminal pose P at time k-1 k-1 The change in the end position of the six-axis industrial robot at time k is ΔP k Add together to get the expected position at time k Complete one iteration; repeat the above steps until the teaching is completed, and calculate P k , P k+1 …P n , where n is the number of sampling cycles during the teaching process;

[0031] Through inverse kinematics solution, a stepping motion is completed. Specifically, the desired posture is known, and the inverse kinematic solution of the six-axis industrial robot is derived according to the Pieper criterion. Then, a stepping motion is completed through the moveit module in ROS.

[0032] Step 3d: Repeat steps 3b and 3c until the operator stops dragging the end of the six-axis industrial robot and the teaching is completed. DETAILED DESCRIPTION

[0033] In order to simplify the teaching process of industrial robots and improve the interactive experience during the teaching process of industrial robots, the present invention aims to provide an industrial robot teaching method based on six-dimensional force sensing information learning.

[0034] To achieve the above object, the technical solution adopted by the present invention mainly includes the following steps:

[0035] Step 1: Equip two robot devices, namely a six-axis industrial robot and a six-axis collaborative robot. The ends of the six-axis industrial robot and the six-axis collaborative robot are both connected to a six-dimensional force sensor;

[0036] Step 2: Establish a neural network model, and obtain the robot terminal motion characteristic library through multiple teaching of the six-axis collaborative robot, and use the motion characteristic library to train the established neural network model. The specific steps are as follows:

[0037] Step 2a, setting the six-axis collaborative robot teaching sampling period T1, and the operator drags the end of the six-axis collaborative robot for teaching. During the teaching process, the joint angle signal of the six-axis collaborative robot and the six-dimensional force information of the end are sampled once in each sampling period until a teaching task is completed;

[0038] The six-axis collaborative robot joint angle signal and the six-dimensional force information of the end are sampled, wherein the six-axis collaborative robot joint angle signal is the six joint angle data of the six-axis collaborative robot, and the six-dimensional force information is the contact force signal of the robot end along the three coordinate axes of x, y, and z and the torque signal around the three coordinate axes of x, y, and z;

[0039] The joint angle signal of the six-axis collaborative robot and the six-dimensional force information of the terminal are sampled. The specific sampling method is to input the joint angle signal of the six-axis collaborative robot and the six-dimensional force information of the terminal through the rosbag module in ROS, and then convert the rosbag file into a txt format;

[0040] Step 2b, repeating step 2a to complete multiple teaching tasks, and obtaining joint angle data and end force sensor data corresponding to multiple teaching tasks of the six-axis collaborative robot;

[0041] Step 2c: pre-process all the obtained force sensor data using Bezier curves to obtain smooth force data and force increment data of each sampling period; and calculate the end position of the six-axis collaborative robot and its end velocity of the corresponding sampling period through forward kinematics;

[0042] As a preference, all the obtained force sensor data are preprocessed using a Bezier curve. Specifically, the force sensor data is substituted into the Bezier curve formula for n times, where n is the quotient of the total teaching time S and the sampling period T1, and the Bezier curve is obtained. Then, the smoothed force data of the i-th sample point after processing is Where i is an integer in the range [0, S / T1];

[0043] As a preferred method, the force increment data calculation method of each sampling period is:

[0044] As a preferred embodiment, the end position of the six-axis collaborative robot is calculated by forward kinematics and described by x, y, z coordinates and zyx Euler angles;

[0045] The terminal position and posture of the six-axis collaborative robot are calculated by forward kinematics: first, the coordinate system relationship between the links of the robot is established, then the homogeneous transformation matrix of the conversion between the coordinate systems is calculated by the DH parameter table, and finally the forward kinematic solution of the robot is derived;

[0046] Step 2d, establishing a robot terminal motion characteristic library, which is composed of the smooth force data obtained in step 2c, the force increment data of each sampling period, and the terminal speed;

[0047] Step 2e, constructing a neural network with the force sensor data and each cycle force increment data as input and the terminal velocity as output; and using the data in the motion characteristic library established in step 2d to train the neural network to obtain the final regression model;

[0048] Preferably, the neural network is a BP neural network, which consists of an input layer, two hidden layers and an output layer: the number of nodes in the input layer is 12, and the corresponding inputs are the contact force signals of the robot end along the three coordinate axes of x, y, and z, the torque signals around the three coordinate axes of x, y, and z, and the increments of each force signal and torque signal relative to the previous sampling period; the number of nodes in the two middle layers is 10; the number of nodes in the output layer is 6, corresponding to the speed of the robot end along the three coordinate axes of x, y, and z and the angular velocity around the three coordinate axes of x, y, and z; the Relu function is selected as the activation function, and the mean square error function is selected as the loss function;

[0049] Step 3: Apply the final regression model obtained through training to the teaching process of the six-axis industrial robot. The specific steps are as follows:

[0050] Step 3a, setting the six-axis industrial robot teaching sampling period T2, and having the operator drag the end of the six-axis industrial robot, while the six-dimensional force sensor connected to the end collects the six-dimensional force information of the end in real time;

[0051] Step 3b: In each sampling period T2, the collected terminal six-dimensional force information f e Processing and generating the end driving force F and the force increment of each sampling period, and inputting them into the final regression model in real time, and outputting the expected speed of the end of the six-axis industrial robot corresponding to each sampling period T2;

[0052] As a preferred method, the collected terminal six-dimensional force information f e Processing and generating the final driving force F d and the force increment of each sampling period. The specific method is to use a piecewise function H(f e ) for the terminal driving force F d Mapping, that is, F d =H(f e ), where H(f e ) is expressed as follows:

[0053]

[0054] Among them, K P is the gain coefficient, and its value range is (0, +∞); K t is the motion threshold, and its value is related to the maximum value of the smoothing force in step 2c.

[0055] The six-dimensional force information f e The purpose of the processing is to prevent excessive external force input from causing the instantaneous speed of the end of the six-axis industrial robot to be too large;

[0056] Step 3c, using the obtained desired velocity of the end of the six-axis industrial robot, solving the desired posture of the end of the six-axis industrial robot, and finally solving it through inverse kinematics to complete a stepping motion;

[0057] As a preferred method, the desired posture of the end of the six-axis industrial robot is solved, specifically, the desired posture at time k is calculated. It is represented as the terminal pose P estimated at time k-1 k-1 The change of the end position of the six-axis industrial robot at time k is ΔP k The sum of the two, as shown in formula (3); and the estimated terminal pose P at time k-1 k-1 , then use the expected posture at time k-1 and the measured pose Z at time k-1 k-1 Characterized as formula (1), where k w is the confidence coefficient, and its value range is [0, 1]. The closer the confidence is to 0, the more reliable the expected posture is, and the closer the confidence is to 1, the more reliable the measured posture is. The change of the terminal posture of the six-axis industrial robot at time k, ΔP k It can be solved by formula (2), where v k With v k-1 are the terminal velocities of the six-axis industrial robot output by the final regression model at time k and time k-1, respectively, and the sampling period T2 is a known quantity; the mathematical model of the above process is as follows:

[0058]

[0059] ΔP k =(v k +v k-1 )·T2 / 2 (2)

[0060]

[0061] In the actual teaching process, first obtain the expected posture at time k-1 and the measured pose Z k-1 , calculate the estimated terminal pose P at time k-1 k-1 ; Then the six-dimensional force sensor signal at the end of the six-axis industrial robot collected at time k is processed to obtain the end driving force F d Then input it into the final regression model to obtain the terminal velocity v of the six-axis industrial robot k , and calculate the change in the end position of the six-axis industrial robot at time k ΔP k ; The estimated terminal pose P at time k-1 k-1 The change in the end position of the six-axis industrial robot at time k is ΔP k Add together to get the expected position at time k Complete one iteration; repeat the above steps until the teaching is completed, and calculate P k , P k+1 …P n , where n is the number of sampling cycles during the teaching process;

[0062] Through inverse kinematics solution, a stepping motion is completed. Specifically, the desired posture is known, and the inverse kinematic solution of the six-axis industrial robot is derived according to the Pieper criterion. Then, a stepping motion is completed through the moveit module in ROS.

[0063] Step 3d: Repeat steps 3b and 3c until the operator stops dragging the end of the six-axis industrial robot and the teaching is completed.

Claims

1. An industrial robot teaching method based on six-dimensional force sensing information learning, characterized in that: The following steps are involved: Step 1: Equip two robot devices, namely a six-axis industrial robot and a six-axis collaborative robot. The ends of the six-axis industrial robot and the six-axis collaborative robot are both connected to a six-dimensional force sensor; Step 2: Establish a neural network model, and obtain the robot terminal motion characteristic library through multiple teaching of the six-axis collaborative robot, and use the motion characteristic library to train the established neural network model. The specific steps are as follows: Step 2a, setting the six-axis collaborative robot teaching sampling period T1, and the operator drags the end of the six-axis collaborative robot for teaching. During the teaching process, the joint angle signal of the six-axis collaborative robot and the six-dimensional force information of the end are sampled once in each sampling period until a teaching task is completed; Step 2b, repeating step 2a to complete multiple teaching tasks, and obtaining joint angle data and end force sensor data corresponding to multiple teaching tasks of the six-axis collaborative robot; Step 2c: pre-process all the obtained force sensor data using Bezier curves to obtain smooth force data and force increment data of each sampling period; and calculate the end position of the six-axis collaborative robot and its end velocity of the corresponding sampling period through forward kinematics; Step 2d, establishing a robot terminal motion characteristic library, which is composed of the smooth force data obtained in step 2c, the force increment data of each sampling period, and the terminal speed; Step 2e, constructing a neural network with the force sensor data and each cycle force increment data as input and the terminal velocity as output; And use the data in the motion characteristic library established in step 2d to train the neural network to obtain the final regression model; Step 3: Apply the final regression model obtained through training to the teaching process of the six-axis industrial robot. The specific steps are as follows: Step 3a, setting the six-axis industrial robot teaching sampling period T2, and having the operator drag the end of the six-axis industrial robot, while the six-dimensional force sensor connected to the end collects the six-dimensional force information of the end in real time; Step 3b: In each sampling period T2, the collected terminal six-dimensional force information f e Processing and generating the final driving force F d and the force increment of each sampling period, and input it into the final regression model in real time, and output the expected speed of the end of the six-axis industrial robot corresponding to each sampling period T2; Step 3c, using the obtained desired velocity of the end of the six-axis industrial robot, solving the desired posture of the end of the six-axis industrial robot, and finally solving it through inverse kinematics to complete a stepping motion; Step 3d: Repeat steps 3b and 3c until the operator stops dragging the end of the six-axis industrial robot and the teaching is completed.

2. An industrial robot teaching method based on six-dimensional force sensing information learning according to claim 1, characterized in that: In step 2c, all the obtained force sensor data are preprocessed using the Bezier curve. The specific method is to substitute the force sensor data into the n-time Bezier curve formula, where n is the quotient of the total teaching process duration S and the sampling period T1, and obtain the Bezier curve. Then the smoothed force data of the i-th sample point after processing is Where i is an integer in the range [0, S / T1].

3. An industrial robot teaching method based on six-dimensional force sensing information learning according to claim 1, characterized in that: In step 3b, the collected terminal six-dimensional force information f e Processing and generating the final driving force F d and the force increment of each sampling period. The specific method is to use a piecewise function H(f e ) for the terminal driving force F d Mapping, that is, F d =H(f e ), where H(f e ) is expressed as follows: Among them, K P is the gain coefficient, and its value range is (0, +∞); K t is the motion threshold, and its value is related to the maximum value of the smoothing force in step 2c.

4. An industrial robot teaching method based on six-dimensional force sensing information learning according to claim 1, characterized in that: In step 3c, the desired posture of the end of the six-axis industrial robot is solved, specifically, the desired posture at time k is calculated. It is represented as the terminal pose P estimated at time k-1 k-1 The change of the end position of the six-axis industrial robot at time k is ΔP k The sum of the two, as shown in formula (3); and the estimated terminal pose P at time k-1 k-1 , then use the expected posture at time k-1 and the measured pose Z at time k-1 k-1 Characterized as formula (1), where k w is the confidence coefficient, and its value range is [0, 1]. The closer the confidence is to 0, the more reliable the expected posture is, and the closer the confidence is to 1, the more reliable the measured posture is. The change of the terminal posture of the six-axis industrial robot at time k, ΔP k It can be solved by formula (2), where v k With v k-1 are the terminal velocities of the six-axis industrial robot output by the final regression model at time k and time k-1, respectively, and the sampling period T2 is a known quantity; the mathematical model of the above process is as follows: ΔP k =(v k +v k -1)·T2 / 2 (2) In the actual teaching process, first obtain the expected posture at time k-1 and the measured pose Z k-1 , calculate the estimated terminal pose P at time k-1 k-1 ; Then the six-dimensional force sensor signal at the end of the six-axis industrial robot collected at time k is processed to obtain the end driving force F d Then input it into the final regression model to obtain the terminal velocity v of the six-axis industrial robot k , calculate the change of the end position of the six-axis industrial robot at time k ΔP k ; The estimated terminal pose P at time k-1 k-1 The change in the end position of the six-axis industrial robot at time k is ΔP k Add together to get the expected position at time k Complete one iteration; repeat the above steps until the teaching is completed, and calculate P k , P k+1 …P n , where n is the number of sampling cycles during the teaching process.

Images