Robot tail end anti-impact compliant control method, device and system and storage medium

Abstract

The invention discloses an anti-impact compliance control method, device and system for a robot tail end and a storage medium. The force information of the robot tail end is estimated based on the position, speed and joint motor torque information of a robot; and the abrupt change torque of the end effector is compensated through the trained network system, and the anti-impact performance and the contact flexibility performance of the robot end effector are improved.

Description

Technical Field

[0001] This invention belongs to the field of robotics technology, and particularly relates to a method, device, system, and storage medium for anti-impact compliance control of a robot end effector. Background Technology

[0002] In modern industrial production, industrial robots play a crucial role in manufacturing. With the rapid rise of high-tech industries, the demand for processing precision has also increased. In some high-precision production processes, industrial robots are gradually replacing humans in production operations. In traditional industrial production, industrial robots achieve specific tasks in industrial scenarios through position control. After receiving a series of instructions from humans, the robot performs its work according to the programmed motion path. Typical applications include material handling, spraying, and welding. However, in applications requiring interaction with the environment, such as grinding, polishing, and assembly, traditional position control is insufficient. Due to processing errors, positioning errors, and robot motion errors, position control alone cannot guarantee the success rate of robot assembly. If a position control system is used alone, any deviation between the robot's position and the designated path will generate significant environmental contact forces, potentially damaging the workpiece. Therefore, the increasing complexity and precision of production requirements place higher demands on the compliant control of robots.

[0003] Currently, the compliance algorithms commonly used in industrial robots for high-compliance assembly problems such as shaft-hole assembly employ the use of force / torque sensors. This aims to solve the precision assembly problems of easily breakable, small-sized, and closely spaced workpieces in industrial settings. The main challenge lies in determining the relationship between contact force and the relative position of the shaft and hole, as well as unknown deviations in the pose between the shaft and hole. The robot's end effector adjusts its pose based on joint torque feedback from torque sensors when it comes into contact with the environment. However, inaccurate and computationally time-consuming dynamic models can lead to errors or delays in the prediction of actual forces and torques, resulting in dangerous situations during assembly, damage to assembled parts, and even damage to the robot itself.

[0004] In summary, robotics technology is now widely used in high-precision and high-complexity manufacturing. However, it commonly suffers from low efficiency, high assembly error rates, and high accident rates when facing complex contact environments. To enable industrial robots to better perform complex and high-precision tasks, and address the problems encountered in industrial assembly scenarios, it is necessary to propose a compliant assembly algorithm that is simple to apply, highly adaptable, and has a wide range of applications. This would allow industrial robots to better adapt to complex assembly scenarios, thereby increasing the degree to which industrial robots replace humans in industrial automation. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, device, system, and storage medium for anti-impact compliance control of robot end effectors.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A robot end effector shock-resistant compliant control method includes: The expected and actual positions, velocities, and accelerations of the motors at each drive joint of the robot are collected using an encoder; the actual drive torque of each drive joint is collected using a current signal acquisition card; a disturbance observer is used instead of a torque sensor as the torque information acquisition device for the end effector. The disturbance observer estimates the abrupt torque information of the end effector by calculating the abrupt torque of the end effector through dynamic equations based on the sudden change in joint torque when the robot's end effector is subjected to force. Calculate the desired joint driving torque; A sample matrix is ​​constructed using the desired position, desired velocity, desired joint driving torque, actual joint motor driving torque, motor ripple torque, and joint friction torque. Using the sample matrix as input and the mutation torque compensation of the end effector as output, an end effector mutation torque compensation prediction network is constructed. The training objective is to minimize the difference between the end effector torque mutation information and the output of the training network. The trained mutation torque compensation prediction network outputs the end effector mutation torque and calculates the compensation value of each joint through the dynamic model to compensate the motor command signal.

[0007] The present invention also provides a robot end effector shock compliant control device, comprising: The first processing module uses an encoder to collect the expected and actual positions, velocities, and accelerations of the motors at each drive joint of the robot; it uses a current signal acquisition card to collect the actual drive torque of each drive joint; and it uses a disturbance observer instead of a torque sensor as the torque information collector for the end effector. The disturbance observer calculates and estimates the abrupt torque information of the end effector based on the sudden change in joint torque when the robot's end effector is subjected to force through dynamic equations. The second processing module is used to calculate the desired joint driving torque; The third processing module constructs a sample matrix using the desired position, desired velocity, desired joint driving torque, actual joint motor driving torque, motor ripple torque, and joint friction torque. The fourth processing module takes the sample matrix as input and the mutation torque compensation of the end effector as output to construct the end effector mutation torque compensation prediction network. It also uses the minimum difference between the end effector torque mutation information and the output of the training network as the training objective to train the mutation torque compensation prediction network. The fifth processing module uses the trained mutation torque compensation prediction network to output the end effector mutation torque and calculates the compensation value of each joint through the dynamic model to compensate the motor command signal.

[0008] The present invention also provides a robot end effector shock compliant control system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a robot end effector shock compliant control method when executed by the processor.

[0009] The present invention also provides a storage medium storing a computer program, which executes a robot end effector anti-impact compliance control method when running.

[0010] This invention estimates the force information of the robot end effector based on the robot's position, velocity, and joint motor torque information; and completes the calibration compensation of the robot end effector contact force estimation algorithm according to the supervised learning method of convolutional neural network, thereby improving the compliance of the end effector contact during assembly. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0012] Figure 1 It is a convolutional matrix structure; where, For the desired positions of each joint, For the desired speed of each joint, Let T be the desired torque for each joint, and T be the actual driving torque for each joint. For the ripple torque of the motor at each joint, For the friction torque of each joint, each column contains the extracted or calculated values ​​for each of the six joints. Figure 2 It is a neural network structure; Figure 3 A flowchart for neural network training; Figure 4 This is a schematic diagram of the anti-impact compliance control of the robot end effector according to the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] Example 1: This invention provides a robot end effector shock-resistant compliance control method. It estimates the force information of the robot's end effector based on the robot's position, velocity, and joint motor torque information, and uses a trained network system to compensate for sudden torque changes in the end effector, thereby improving the shock resistance and contact compliance performance of the robot's end effector. The method includes the following steps: Step 1: Data Collection For a robot model to execute a specified task trajectory, an encoder is used to collect the expected and actual positions, velocities, and accelerations of the motors at each drive joint of the robot; a current signal acquisition card is used to collect the actual drive torque of each drive joint; a disturbance observer is used instead of a torque sensor as the torque information acquisition device for the end effector. The disturbance observer estimates the abrupt torque information of the end effector by calculating the abrupt torque of the end effector through dynamic equations based on the sudden change in joint torque when the robot end effector is subjected to force. ; in, For sudden change torque, For functions of the Lagrange equation, For generalized coordinates, For generalized speed.

[0016] Step 2: Calculate the desired torque Using the desired position, velocity, and acceleration collected in step one, the desired driving torque is calculated using the robot's rigid body dynamics model. ; in, For the desired driving torque, For the robot's inertia matrix, The matrix of centrifugal force and Kirchhoff force for the robot. For the gravity matrix, For generalized coordinates, For generalized speed, This refers to generalized acceleration.

[0017] Step 3: Construct the sample matrix For the expected and actual positions and velocities of each joint obtained in steps one and two, interpolation sampling is performed, and the sampled values ​​are used as matrix elements to construct a matrix. Simultaneously, because a significant motor ripple torque was found in the joint torques, this motor ripple torque was extracted as a matrix element for training. Therefore, the corrected dynamic equation can be simplified as follows: ; in, The motor's ripple torque is abbreviated as . ; This represents the joint friction torque, abbreviated as . Assume that their variation patterns are related to joint position and joint velocity. When the joint rotates at a low, uniform speed, assume... and Approximately 0, can be considered and It is also approximately 0. Assuming that the gravitational torque term and the motor ripple torque term are only related to the joint position and not to the direction of rotation; and that the magnitudes of the joint friction torques are equal but their signs are opposite during forward and reverse rotation, then the joint friction torque and motor ripple torque calculated from experimental data for a given desired speed can be obtained as follows: ; in, This represents the positive torque after resampling; Indicates the negative torque after resampling; This represents the theoretical joint torque calculated using the expression for the gravitational moment term; and This represents the joint friction torque and motor ripple torque calculated from the forward and reverse rotational torques.

[0018] like Figure 1 As shown, a 6×6 matrix is ​​formed by the desired position, desired speed, desired joint driving torque, actual driving torque of the joint motor, motor ripple torque, and joint friction torque. The corresponding six-degree-of-freedom torque error compensation values ​​of the end effector are calibrated. The 6×6 matrix is ​​then subjected to row and column transformations to form a sample matrix to prevent overfitting of the neural network during training, which would cause a decrease in accuracy. The above two types of matrices are used as training sample data for future reference.

[0019] Step 4: Construct a sudden change moment compensation prediction network like Figure 2 As shown, the mutation moment compensation prediction network consists of an input layer, a first convolutional layer, a second convolutional layer, a nonlinear activation function layer, a pooling layer, a first fully connected layer, and a second fully connected layer, connected in sequence. Each layer is defined as follows: Input layer: The input layer has a two-input structure, with two 6×6 matrices as inputs (the 6×6 matrix extracted in step three and the sample matrix obtained by its row and column transformations). Convolutional layers: The input layer is followed by a first convolutional layer and a second convolutional layer. Both the first and second convolutional layers have 3×3 kernels. The first convolutional layer has 32 sets of kernels, with 2 kernels per set. The second convolutional layer has 64 sets of kernels, with 32 kernels per set. The input to the convolutional layers is padded with zeros. The 3×3 convolutional kernels are used to extract features from the robot's joint variable matrix, with a stride of 1 in both the horizontal and vertical directions.

[0020] Non-linear activation function layer: A non-linear activation function layer is cascaded after two convolutional layers. Pooling layer: An activation function layer is followed by a pooling layer, and the size of each feature layer remains 6×6. Since the input layer size is very small, no merging operation is performed.

[0021] Fully connected layer: Following this are two cascaded fully connected layers, the first and second, with 2304 and 512 neurons respectively. The CNN output is 6×1, with elements corresponding to the predicted compensation values ​​for the abrupt torque error of the robot's end effector.

[0022] Step 5: Train the network like Figure 3 As shown, firstly, the expected and actual positions, velocities, and accelerations of each joint are interpolated and sampled. The sampled values ​​are used as matrix elements to construct a matrix. The matrix consists of the expected position, expected velocity, expected acceleration, expected driving torque, actual driving torque, and joint friction torque of the robot's six joint drive motors, forming a 6×6 matrix. Then, through the aforementioned input layer, convolutional layer, nonlinear activation function layer, pooling layer, and fully connected layer, a sudden torque compensation prediction network is constructed. Finally, the network is trained with the goal of minimizing the difference between the output of the convolutional neural network and the end effector torque calculated in step one.

[0023] Step Six: Implement Compensation like Figure 4 As shown, the network is integrated into the control system of each joint of the robot, and the sudden torque compensation network is integrated into the control system of each joint of the robot. The desired position, desired speed, desired torque, actual driving torque, motor ripple torque and friction torque of the robot are compared with the input and actual output values ​​of the robot to obtain the error value. The error value obtained by the sudden torque compensation prediction network is then compensated on the motor command position signal by the Jacobian matrix, thereby improving the output accuracy.

[0024] Example 2: This invention also provides a robot end effector shock compliant control device, comprising: The first processing module uses an encoder to collect the expected and actual positions, velocities, and accelerations of the motors at each drive joint of the robot; it uses a current signal acquisition card to collect the actual drive torque of each drive joint; and it uses a disturbance observer instead of a torque sensor as the torque information collector for the end effector. The disturbance observer calculates and estimates the abrupt torque information of the end effector based on the sudden change in joint torque when the robot's end effector is subjected to force through dynamic equations. The second processing module is used to calculate the desired joint driving torque; The third processing module constructs a sample matrix using the desired position, desired velocity, desired joint driving torque, actual joint motor driving torque, motor ripple torque, and joint friction torque. The fourth processing module takes the sample matrix as input and the mutation torque compensation of the end effector as output to construct the end effector mutation torque compensation prediction network. It also uses the minimum difference between the end effector torque mutation information and the output of the training network as the training objective to train the mutation torque compensation prediction network. The fifth processing module uses the trained mutation torque compensation prediction network to output the end effector mutation torque and calculates the compensation value of each joint through the dynamic model to compensate the motor command signal.

[0025] Example 3: This invention also provides a robot end effector shock compliant control system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a robot end effector shock compliant control method when executed by the processor.

[0026] Example 4: This invention also provides a storage medium storing a computer program, which executes a robot end effector anti-impact compliant control method during runtime.

[0027] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A robot end impact-avoiding compliance control method, characterized by, The method comprises the following steps: acquiring the expected and actual positions, speeds and accelerations of the motors of the driving joints of the robot by using an encoder; acquiring the actual driving torques of the driving joints by using a current signal acquisition card; using a disturbance observer to replace a torque sensor as a torque information collector of the end effector, the disturbance observer calculating and estimating the sudden torque information of the end effector according to the sudden change of the joint torque when the end effector of the robot is subjected to force; ； wherein, is the mutation torque, is the Lagrange equation function, is the generalized coordinate, is the generalized velocity; calculating the expected joint driving torque; calculating the expected driving torque by means of the rigid body dynamics model of the robot with the acquired expected positions, speeds and accelerations; ； wherein, is a desired driving torque, is a robot inertia matrix, is a robot centrifugal and Coriolis force matrix, is a gravity matrix, is a generalized coordinate, is a generalized velocity, is a generalized acceleration; constructing a sample matrix; constructing a 6*6 matrix with the expected positions, speeds, expected joint driving torque, actual joint motor driving torque, motor fluctuation torque and joint friction torque, calibrating the corresponding six-degree-of-freedom torque error compensation values of the end effector, and performing row transformation and column transformation on the constructed 6*6 matrix to form a sample matrix; constructing an end effector sudden torque compensation prediction network with the sample matrix as the input and the sudden torque compensation of the end effector as the output, the sudden torque compensation prediction network being composed of an input layer, a first convolutional layer, a second convolutional layer, a nonlinear activation function layer, a pooling layer, a first full connection layer and a second full connection layer connected in sequence; wherein the convolution kernel sizes of the first convolutional layer and the second convolutional layer are both 3*3, the first convolutional layer has 32 groups of convolution kernels, each group having 2 kernels, and the second convolutional layer has 64 groups of convolution kernels, each group having 32 kernels; and the minimum difference between the sudden torque information of the end effector and the output of the trained network is used as the training target to train the sudden torque compensation prediction network; compensating the motor command signal with the joint compensation values calculated by the dynamics model with the sudden torque of the end effector output by the trained sudden torque compensation prediction network.

2. A robot end impact protection compliance control device that implements the robot end impact protection compliance control method according to claim 1, characterized by, The method comprises the following steps: a first processing module acquires the expected and actual positions, speeds and accelerations of the motors of the driving joints of the robot by using an encoder; a current signal acquisition card is used to acquire the actual driving torques of the driving joints; a disturbance observer is used to replace a torque sensor as a torque information collector of the end effector, the disturbance observer calculating and estimating the sudden torque information of the end effector according to the sudden change of the joint torque when the end effector of the robot is subjected to force; a second processing module is used to calculate the expected joint driving torque; a third processing module is used to construct a sample matrix with the expected positions, speeds, expected joint driving torque, actual joint motor driving torque, motor fluctuation torque and joint friction torque; a fourth processing module is used to construct an end effector sudden torque compensation prediction network with the sample matrix as the input and the sudden torque compensation of the end effector as the output, and the minimum difference between the sudden torque information of the end effector and the output of the trained network is used as the training target to train the sudden torque compensation prediction network; a fifth processing module is used to compensate the motor command signal with the joint compensation values calculated by the dynamics model with the sudden torque of the end effector output by the trained sudden torque compensation prediction network.

3. A robot end impact-avoiding compliance control system, characterized by, The method comprises the following steps: A memory and a processor, the memory having stored thereon a computer program run by the processor, the computer program, when run by the processor, performing the robot end impact-avoiding compliance control method of claim 1.

4. A storage medium, characterized by The storage medium has stored thereon a computer program, the computer program, when run, performing the robot end impact-avoiding compliance control method of claim 1.

Images