Control method for mechanical arm of power transmission line hot-line work robot

Abstract

The invention designs a mechanical arm control method for a power transmission line hot-line work robot. The mechanical arm control method comprises the steps that S1, a real-time mechanical arm kinematics model is established through a current angle; and S2, changing to a base coordinate and obtaining an actual movement speed through impedance control. And S3, calculating the dynamic speed to obtain the expected pose of the robot. And S4, subtracting the pose from the actual pose to obtain the running time, obtaining the correction speed, and calculating the joint speed increment in combination with a Jacobian matrix. And S5, repeating the steps 1)-4) until a convergence condition is met. And S6, the current joint angle is output, and the current posture of the mechanical arm is obtained. According to the method, the weighted generalized inverse matrix of the Jacobian matrix is combined with the tail end speed of the mechanical arm in impedance control, the angle increment of the ultralimit joint is reduced, compliant control over the multi-degree-of-freedom mechanical arm is achieved, the operation target recognition technology is optimized by improving the YOLOv4 algorithm, and the control precision of the multi-degree-of-freedom mechanical arm is improved. The problem of compliance of contact between the mechanical arm and an unknown environment or object is solved.

Description

[0001] The present invention relates to the fields of robot technology and live working robotic arms, and particularly to a control method for a live working robotic arm of a transmission line. Background Art

[0002] With the rapid development of robot technology, people have higher requirements for the compliant and safe human-robot collaboration ability of robots. Therefore, robotic arms with a redundant structure for compliant control have gradually come into people's view. Compared with traditional robotic arms, redundant robotic arms have better flexibility, fault tolerance, and stronger collaboration ability, thus having a wider application field. When it comes to solving the compliance problem of the robotic arm in contact with an unknown environment or object, the robot compliant control technology for human-robot interaction is particularly crucial.

[0003] Therefore, for the live working robotic arm of the present invention, a robotic arm compliant control technology and an operation area recognition algorithm based on joint priority are proposed. The current methods for realizing robotic arm compliant control include the following several types:

[0004] 1) Using impedance control, a force output is generated according to the input of the position deviation. In this control mode, the control system of the robot will calculate the reference force required to achieve the desired impedance model, and then make the actual interaction force between the robot and the environment track this desired contact force through the inner-loop force controller.

[0005] 2) Using admittance control, admittance control is a position controller based on force feedback. In admittance control, when the robot is subjected to an external force, the control system will calculate a desired position, and then input this desired position into the inner-loop position control to make the dynamic relationship between the robot and the environment present as the target impedance model. Admittance control requires giving the desired contact force between the robot and the environment, and can be further extended to take the derivative of the calculated desired position to obtain the desired velocity and desired acceleration, so as to drive the robot to move.

[0006] The existing methods for robotic arm compliant control have the following disadvantages:

[0007] 1. Using impedance control, its control performance highly depends on an accurate robotic arm dynamics model and is not universal; in some cases, impedance control may also cause the system to be unstable; at the same time, its complex algorithms and controllers increase the difficulty of system design and debugging.

[0008] 2. When using admittance control, when the robot comes into contact with a rigid body, admittance control often becomes unstable; admittance control has high requirements for sensors and requires a force sensor to detect the contact force. If the force sensor detects a sudden large impulse force, it may cause the divergence of the desired position and speed, resulting in distorted results. Summary of the Invention

[0009] The object of the present invention is to provide a control method for the manipulator of a live working robot for transmission lines, introducing joint priorities to improve the motion efficiency and compliance of a manipulator with multiple degrees of freedom. The application scenarios of the invention are as Figure 1 shown.

[0010] To achieve the above object, the present invention provides the following solution:

[0011] As Figure 2 shown, a control method for the manipulator of a live working robot for transmission lines provided by the present invention introduces a manipulator joint priority mechanism to achieve compliant control of the manipulator of the live working robot for transmission lines, including the following steps:

[0012] 1) Establish a real-time manipulator kinematic model through the current angles;

[0013] 2) Transform the contact force to the base coordinate and obtain the actual motion speed through impedance control;

[0014] 3) Calculate the desired pose of the robot according to the motion speed;

[0015] 4) Subtract the actual pose from the desired pose, divide by the running time to obtain the correction speed, and calculate the joint speed increment in combination with the Jacobian matrix;

[0016] 5) Repeat steps 1)-4) until the convergence condition is met;

[0017] 6) Output the current joint angles to obtain the current manipulator pose.

[0018] Furthermore, by introducing joint priorities, some joints of the manipulator with multiple degrees of freedom move less or not at all, achieving the effect when all joints are involved in the motion, thereby restricting the motion of some joints; generally speaking, joints closer to the base or joints with larger loads should be set with lower joint priorities to reduce their total amount of motion, which can prevent some joints from reaching the limit range of motion during the robot's movement and keep each joint within its working range; to some extent, setting priorities in this way can also reduce the torque required for the end effector to move, thereby improving the mechanical performance of the manipulator.

[0019] Furthermore, for the live working ROI area, the CSPDarknet53 backbone network is used to initially extract feature information from the original input image. CSPDarknet53 is based on the previous generation Darknet53 and incorporates the CrossStagePartialNetwork (CSPNet) structure in 2019 to improve the activation function of the network. At the same time, the convolution block and residual network are modified to increase the convolution speed, resulting in the Backbone residual network structure. To reduce the number of parameters in the network model and simplify FLOPS, the repeated gradient problem in the convolution network is improved, enhancing the monitoring accuracy and convolution speed.

[0020] Furthermore, for the enhanced feature extraction network, pooling operations are performed by the maximum pooling method. The sizes of the SPP pooling boxes are 5, 9, and 13 respectively. After pooling operations with different numbers of parameters, the image features are stacked and output. The PANet network is composed of upsampling, downsampling, and depth convolution. The upsampling and downsampling are divided into two parts: FPN and PAN. Among them, for the input feature information, the low-resolution features are processed by FPN, and then combined with PAN to perform the downsampling transfer process after upsampling, improving the accuracy of sampling and noise robustness.

[0021] Furthermore, based on the YOLOv4 network, depthwise separable convolutions are used instead of ordinary convolutions, and an improved attention mechanism is added to select the spine feature points, which not only ensures the inference speed and accuracy but also reduces the model size and improves the accuracy of identifying the operation target during the operation of the robotic arm.

[0022] The present invention overcomes the deficiencies of the prior art and combines the weighted generalized inverse matrix of the Jacobian matrix with the end velocity of the robotic arm in impedance control to reduce the angular increment of the over-limit joints, achieving the compliant control of the multi-degree-of-freedom robotic arm. It also optimizes the operation target recognition technology by improving the YOLOv4 algorithm, solving the compliance problem of the robotic arm in contact with unknown environments or objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is the application scenario of the present invention;

[0024] Figure 2 is the physical model diagram of the robot impedance control;

[0025] Figure 3 is the flowchart of the robotic arm compliant control algorithm under joint priority provided by the present application;

[0026] Figure 4 is the network structure diagram of the improved YOLOV4 algorithm provided by the present application. DETAILED DESCRIPTION OF THE INVENTION

[0027] The technical solutions in the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

[0028] The purpose of the present invention is to provide a compliant control method for the robotic arm in live working, introducing joint priorities to improve the motion efficiency and compliance of the robotic arm with multiple degrees of freedom.

[0029] Different from the existing compliant control methods for robotic arms, the present invention introduces joint priorities, enabling some joints of the robotic arm with multiple degrees of freedom to move less or not at all, achieving the effect when all joints participate in the movement, and thus restricting the movement of some joints. Generally speaking, the joints closer to the base or the joints with larger loads should be set with smaller joint priorities to reduce their total amount of movement, which can avoid some joints reaching the limit range of motion during the movement of the robot and enable each joint to be within its working range. To some extent, setting priorities in this way can also reduce the torque required for the movement of the end effector, thereby improving the mechanical performance of the robotic arm.

[0030] The path planning method for the substation oil sample collection robot includes the following contents:

[0031] The Jacobian matrix represents the differential relationship between the end effector of the robot and each joint, and is the general transmission ratio for transmitting the motion speed from the robot joint space to the operation space. Assuming X = [x1, x2,..., xn] represents the end pose of the robot, and at this time the joint angles of the robotic arm are q = [q1, q2, q3, q4, q5, q6, q7, q8], then the relationship between X and q is

[0032] X = T(q)

[0033] Differentiating both sides of the formula simultaneously gives:

[0034]

[0035] Where: J(q) is the Jacobian matrix of the robotic arm; represents the end velocity vector of the robotic arm; is the joint space velocity vector.

[0036] For the robotic arm in any pose, its end pose can be reduced to 3 displacements in the directions of the base coordinate axes and 3 rotations around the base coordinate axes, and it is written in the form of a 6×1 matrix. The first three rows represent the position of the robotic arm actuator in the base coordinate, and the last three rows represent the rotation of the robotic arm actuator around 3 coordinate axes. Differentiating this matrix can obtain the translational velocity v and rotational velocity ω of the end effector relative to the base coordinate. At this time, the motion equation can be written as:

[0037]

[0038] where v = [v x , v y , v z T is the moving speed of the end effector along the corresponding axis of the base coordinate; ω = [ω x , ω y , ω z T is the angular velocity of the end effector rotating around the base coordinate axis.

[0039] For a traditional six-degree-of-freedom robot, the size of its Jacobian matrix is 6×6, and the formula can be directly used for calculation. However, in reality, the manipulators encountered cannot guarantee other situations, and in this case, the generalized inverse matrix of the Jacobian matrix is required. The generalized inverse matrix of the Jacobian matrix is divided into two cases: the degree of freedom of the manipulator is greater than 6 and the degree of freedom is less than 6. The mathematical expressions of the generalized inverse matrix of the Jacobian matrix of manipulators with different degrees of freedom are as follows:

[0040]

[0041] Figure 2 is the physical model of the impedance control of the robot. For the impedance control of the manipulator, impedance represents the relationship between the contact force at the end of the robot and the difference between the desired trajectory and the actual trajectory. The three most commonly used mathematical expression forms of impedance control are

[0042]

[0043] where: M d , B d , K d respectively represent the inertia matrix, damping matrix, and stiffness matrix in the desired impedance model; X, respectively represent the actual displacement, velocity, and acceleration at the end of the manipulator; X d , respectively represent the desired position, desired velocity, and desired acceleration at the end of the manipulator; F e represents the contact force between the end of the manipulator and the environment. When the outside interacts with the end of the manipulator, before the manipulator reaches a stable state, the contact force F e will cause the end of the manipulator to move along the direction of the contact force at a certain speed and acceleration until the manipulator reaches a stable state.

[0044] ​​Introducing joint priorities can enable a robotic arm with multiple degrees of freedom to move less or not at all at certain joints, achieving the same effect as when all joints are involved in the movement, thereby restricting the movement of certain joints. Generally speaking, joints closer to the base or joints with a larger load should be set with lower joint priorities to reduce their total amount of movement. This can prevent certain joints from reaching their movement limit ranges during the robot's movement and ensure that each joint remains within its working range. To some extent, setting priorities in this way can also reduce the torque required for the end effector to move, thereby improving the mechanical performance of the robotic arm.

[0045] Assume that q0 = [q1, q2, q3, q4, q5, q6, q7, q8] is the initial angle of the robotic arm, and the corresponding pose of the end of the robotic arm at this time The change in the end velocity of the robotic arm caused by the external contact force after impedance control is dX, and the expected pose E is calculated based on dX a ′ change:

[0046] E a ′ = E a + dX·dt

[0047] Subtract the calculated pose E from E a ′ and divide by the running time dt to obtain the corrected velocity dX': a0

[0048] dX′ = (E a ′ - E a0 ) / dt

[0049] The Jacobian matrix J of the redundant robotic arm is obtained through MATLAB, and its generalized inverse matrix J is further obtained + :

[0050] J + = J T (J T J) -1

[0051] After obtaining the generalized inverse matrix of the Jacobian matrix, considering the actual movement of each joint, priorities are set for some joints. Thus, a weighted generalized inverse matrix is introduced, and its specific mathematical expression is

[0052]

[0053] where: The weight matrix W is a diagonal matrix. For an eight-degree-of-freedom robotic arm, some joints can be not involved or limitedly involved during the movement. In the weight matrix, the weight ω corresponding to joint i i The larger it is, the lower the priority of this joint. When 1 / ω i ​When the value approaches 0, the corresponding joint priority ki also approaches 0, and then this joint does not participate in the movement of the robotic arm. At this time, the joint angular velocity of the robotic arm is

[0054]

[0055] where dq' is the change in joint angular velocity after correction.

[0056] Update the joint angles of the robotic arm from the joint angular velocity:

[0057] q n = q b + dq′·dt

[0058] where: q is the joint angle; dt represents the movement time of each step of the robotic arm during movement.

[0059] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes, but is not limited to, the embodiments described in the specific embodiments. Any other embodiments obtained by those skilled in the art based on the technical solutions of the present invention also fall within the scope of protection of the present invention.

Claims

1. A control method for the manipulator of a live working robot for transmission lines, which introduces a manipulator joint priority mechanism to achieve compliant control of the manipulator of the live working robot for transmission lines, including the following steps: 1) Establish a real-time manipulator kinematic model through the current angle; 2) Transform the contact force to the base coordinate and obtain the actual motion speed through impedance control; 3) Calculate the desired pose of the robot according to the motion speed; 4) Subtract the desired pose from the actual pose, divide by the running time to obtain the correction speed, and calculate the joint speed increment in combination with the Jacobian matrix; 5) Repeat steps 1)-4) until the convergence condition is met; 6) Output the current joint angle to obtain the current manipulator pose.

2. The manipulator control method of a live working robot for transmission lines according to claim 1, wherein , By introducing joint priorities, the multi-degree-of-freedom manipulator can have some joints move less or not at all, achieving the effect when all joints are involved in the movement, and thus restricting the movement of some joints; generally speaking, the joints closer to the base or the joints with larger loads should be set with smaller joint priorities to reduce their total amount of movement, which can prevent some joints from reaching the limit range of motion during the movement of the robot and keep each joint within its working range; to some extent, setting priorities in this way can also reduce the torque required for the movement of the end effector, thereby improving the mechanical performance of the manipulator.

3. A manipulator control method for a live working robot of a transmission line according to claim 1, characterized in that , For the live working ROI area, initially extract feature information from the original input image through the CSPDarknet53 backbone network; CSPDarknet53 is based on the previous generation Darknet53, drawing on the cross-stage partial network structure CSPNet in 2019 to improve the activation function of the network; at the same time, to improve the convolution speed, the convolution block and the residual network are modified to generate the Backbone residual network structure; to reduce the number of parameters in the network model and simplify FLOPS, the problem of repeated gradients in the convolution network is improved, enhancing the monitoring accuracy and convolution speed.

4. A manipulator control method for a live working robot of a transmission line according to claim 1, characterized in that , For the enhanced feature extraction network, pooling operations are performed in the way of maximum pooling; the sizes of the SPP pooling boxes are 5, 9, and 13 respectively. After pooling operations with different numbers of parameters, the image features are stacked and output; the PANet network is composed of upsampling, downsampling, and depth convolution. The upsampling and downsampling are divided into two parts: FPN and PAN; among them, for the input feature information, the low-resolution features are processed by FPN, and then combined with PAN to perform the transfer processing of downsampling after upsampling, improving the accuracy of sampling and the noise robustness.

5. A manipulator control method for a live working robot of a transmission line according to claim 1, characterized in that , Based on the YOLOv4 network, depthwise separable convolution is used to replace ordinary convolution, and an improved attention mechanism is added for the selection of spine feature points, which not only ensures the inference speed and accuracy but also reduces the model size, improving the accuracy of identifying the operation target during the operation of the manipulator.

Images