A robot kinematics parameter calibration method based on visual multi-point pose constraint

Abstract

The present application relates to a kind of robot kinematics parameter calibration method based on visual multi-point pose constraint, comprising: control end installation camera's robot to be calibrated around the center of calibration board motion, collect robot model data and the nominal pose data of calibration board in multiple pose state during motion;Using hand-eye calibration algorithm obtains initial hand-eye conversion matrix as iteration initial value;Multiple calibration points on calibration board are combined to construct and generate multiple-point pose constraint equation group from the combination of differential pose constraint equation group;The differential pose constraint equation group of single calibration point is the deviation of the nominal pose and actual pose of this calibration point and the relationship of robot kinematics parameter deviation;Multiple-point pose constraint equation group is solved iteratively using weighted least square method, to achieve the robot kinematics parameter deviation obtained when iteration stop relationship is carried out to the kinematics parameter of the robot to be calibrated;It is low in cost, equipment installation precision requirement is low, and actual position of calibration board is not required in calibration process.

Description

Technical Field

[0001] The present invention relates to the field of robots, and in particular to a robot kinematic parameter calibration method based on visual multi-point posture constraints. Background Art

[0002] In current manufacturing processes, traditional robots not only need to perform simple repetitive movements based on a teach-and-play model, but also require offline programming for complex trajectory movements to achieve flexible production of small batches and multiple varieties, as well as multi-robot collaboration. These tasks are particularly difficult for industrial robots, which have high repeatability but poor absolute positioning accuracy. Part errors during the manufacturing and assembly process, deformation under continuous high loads, collisions, wear and tear from long-term use, and joint replacement and repair can all lead to deviations between the actual kinematic parameters and the nominal parameters, resulting in absolute positioning errors at the robot's end effector. In industrial applications, the robot's absolute positioning accuracy is generally roughly tested using the 6-degree-of-freedom feature of the 6R robot end using point-of-sight accuracy.

[0003] Among the factors affecting a robot's absolute positioning accuracy, deviations in kinematic parameters account for approximately 80% of the total error. Therefore, kinematic calibration is the most cost-effective and effective method for improving a robot's absolute positioning accuracy, provided the robot's repeatability is sufficient. Numerous researchers have conducted diverse research and proposed various calibration methods. Robot calibration methods can be categorized as open-loop and closed-loop.

[0004] Open-loop calibration measures the actual position of the robot's end effector based on a world coordinate system. Common measurement equipment used for absolute calibration includes laser trackers, ballbars, and coordinate measuring machines. Precision measurement equipment can provide high-precision measurement results, but the equipment is expensive and requires skilled operators. T. Sun, Filion, and others have proposed open-loop calibration methods for robots based on high-precision external measurement equipment. These methods can measure the robot's end position with high accuracy, identify the robot's joint parameters, and obtain good calibration results. However, the related equipment is bulky and expensive.

[0005] Closed-loop calibration measures the relative pose / position of the end effector, imposing certain constraints such as plane, distance, single-point, and spherical constraints without requiring actual position measurements. While the parameter identification process in closed-loop calibration is challenging, the measurement equipment used is relatively inexpensive. The main drawbacks of most closed-loop calibration methods are the need for large calibration artifacts, time consumption, and low accuracy.

[0006] Joubair achieved kinematic calibration for a 6R robot by establishing plane constraints using a contact probe mounted on the end of the robot and a fixed high-precision flat plate. However, this required the use of a large, high-precision machined standard object, and the contact measurement process was cumbersome, complex, and time-consuming. Wang used machine vision methods to establish point and distance constraints at the end of the robot using a standard sphere, achieving a closed-loop kinematic calibration method for the 6R robot. However, this method required a higher-pixel vision sensor to achieve a measurement accuracy of 0.02mm and had a smaller measurement range.

[0007] Currently, closed-loop calibration methods are not widely used. Several robot manufacturers and other organizations offer one-time open-loop calibration services. However, this is only feasible for customers with one or a few robots, and deploying measurement equipment in complex factory environments is difficult and time-consuming. Cost-effective closed-loop calibration methods have great potential for widespread adoption.

[0008] Cameras are commonly used closed-loop calibration measurement devices. Chen H used the PnP algorithm to achieve high-precision monocular pose measurement. Lee used a neural network to achieve online calibration of the robot's hand-eye relationship based on monocular pose measurement. Sarabandi used a hand-eye calibration algorithm to achieve parameter calibration of the robot's hand-eye relationship. However, the hand-eye calibration does not consider the robot's kinematic error, which will affect the hand-eye calibration accuracy. Summary of the Invention

[0009] In response to the technical problems existing in the prior art, the present invention provides a robot kinematic parameter calibration method based on visual multi-point posture constraints, which can realize the identification and compensation of joint zero position error, connecting rod length error, and hand-eye conversion relationship, without measuring the actual position of the calibration plate during the process.

[0010] According to a first aspect of the present invention, a robot kinematic parameter calibration method based on visual multi-point pose constraints is provided, comprising:

[0011] Step 1: Control the robot to be calibrated with a camera installed at the end to move around the center of the calibration plate, and collect robot model data in multiple posture states during the movement and the nominal posture data of the calibration plate;

[0012] Step 2: Use the hand-eye calibration algorithm to obtain the initial hand-eye conversion matrix as the initial value of the iteration;

[0013] Step 3, generating a multi-point pose constraint equation group by combining the differential pose constraint equation groups of multiple calibration points on the calibration plate; the differential pose constraint equation group of a single calibration point is the relationship between the deviation between the nominal pose and the actual pose of the calibration point and the deviation of the robot kinematic parameters;

[0014] Step 4: Use the weighted least squares method to iteratively solve the multi-point pose constraint equations to calibrate the kinematic parameters of the robot to be calibrated using the robot kinematic parameter deviation obtained when the iteration stops.

[0015] On the basis of the above technical solution, the present invention can also make the following improvements.

[0016] Optionally, the robot model data collected in step 1 includes robot parameters and joint angles;

[0017] The nominal pose data of the calibration plate is obtained based on the visual pose measurement result of the robot.

[0018] Optionally, the process of obtaining the differential pose constraint equations for a single calibration point includes:

[0019] Step 301: Calculate the deviation between the nominal pose and the actual pose as the error model formula: ΔP = P Ab -P nb =f(w+Δw)-f(w);

[0020] Among them, ΔP represents the deviation; P Ab represents the actual posture, i P Ab =f(w+Δw) represents the actual position of point i on the calibration plate; P nb represents the nominal pose, i P nb =f(w) represents the nominal pose of point i on the calibration plate; the subscript n represents n poses, w and Δw are the identifiable parameters of the robot model and their variations, respectively;

[0021] Step 302: Simplify the error model formula using a first-order Taylor expansion and write it as: ΔP = JΔw; J is the posture Jacobian matrix;

[0022] Step 303: Establish a single-point differential pose constraint equation set as follows: Abbreviated as i HΔw= i O;

[0023] Step 304: Generate a multi-point pose constraint equation group by combining the single-point differential pose constraint equation groups, which is expressed as

[0024] Abbreviated as KΔw=B.

[0025] Optionally, the robot model is a DH parameter model;

[0026] Identifiable parameters of the robot model:

[0027] w=[θ2,θ3,θ4,θ5,a2,a3,d3,R e0 ,R e1 ,R e2 ,R e3 ,T e0 ,T e1 ,T e2 ];

[0028] Among them, θ2, θ3, θ4 and θ5 are the correctable joint zero position parameters, a2, a3, d3 are the correctable link length parameters of the robot, [R e0 ,R e1 ,R e2 ,R e3 ] and [T e0 ,T e1 ,T e2 ] is the hand-eye relationship parameter.

[0029] Optional, nominal pose of the calibration points on the calibration board in, and The transformation matrix of adjacent joint links in the robot model Get; wherein, i=1, 2, 3, 4, 5 or 6, when the robot is a 6-degree-of-freedom serial link robot, base=0, end=6; s represents a sine function; c represents a cosine function;

[0030] is the transformation matrix from the camera coordinate system to the robot end coordinate system, is the transformation matrix from the calibration plate coordinate system to the camera coordinate system.

[0031] Optionally, in step 302, the error model formula is simplified by a first-order Taylor expansion to:

[0032]

[0033] Optionally, the process of establishing the single-point differential pose constraint equation group in step 303 is:

[0034] Based on the nominal pose data of the calibration plate obtained by visual pose measurement at different robot poses, a single-point pose constraint equation group is established:

[0035] Abbreviated as i HΔw= i O;

[0036] Eliminate the actual pose of point i in the pose constraint equations by using the difference method i P Ab , and obtain the single-point differential pose constraint equations.

[0037] The multi-point pose constraint equations are generated by combining the single-point differential pose constraint equations and are expressed as Abbreviated as KΔw=B.

[0038] Optionally, using the weighted least squares method in step 4 to solve the multi-point pose constraint equation group KΔw=B is equivalent to solving (K T K+λI)Δw=K T B equation; where λ is the weight, and the weight λ is passed through B T The changing trend of B is optimized with variable step size.

[0039] Optionally, the condition for stopping iteration in step 4 includes: satisfying the maximum number of iterations or the terminal winding point error;

[0040] The calculation process of the terminal winding point error includes: measuring m points in multiple postures for n calibration points n The average of the position errors of each pose relative to the point is taken as the terminal point error.

[0041] The present invention provides a robot kinematic parameter calibration method based on visual multi-point posture constraints. First, the camera in the common robot hand-eye system can be directly used, which is low-cost. It can be used in conjunction with a low-cost calibration plate to complete kinematic calibration. The equipment installation accuracy requirements are low, and it is easy to install, disassemble and carry. The actual position of the calibration plate does not need to be measured during the calibration process. Secondly, the high-precision calibration plate target provides the basis for high-precision measurement. The sub-pixel feature extraction algorithm of the calibration plate target and the monocular vision posture method further improve the measurement accuracy and stability of the system. Then, the system calibration process is carried out at the robot workstation without the need to disassemble or transport the robot. There are no special requirements for the workspace required for calibration. At the same time, since the calibration tool is low-cost and easy to install, long-term on-site precision monitoring can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of a robot kinematic parameter calibration method based on visual multi-point pose constraints provided by an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of the visual pose measurement target and feature point extraction results provided by an embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of the HSR-JR605 robot hand-eye system;

[0045] Figure 4 This is a schematic diagram of the coordinate system of each link of the HSR-JR605 robot;

[0046] Figure 5A schematic diagram of an experimental system for an application embodiment of a robot kinematic parameter calibration method based on visual multi-point pose constraints provided by the present invention;

[0047] Figure 6 The end position error of an application embodiment of a robot kinematic parameter calibration method based on visual multi-point pose constraints provided by the present invention;

[0048] Figure 7 The present invention provides an application embodiment of a robot kinematic parameter calibration method based on visual multi-point posture constraints to compensate for the front and rear end position error distribution histogram. DETAILED DESCRIPTION

[0049] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0050] Figure 1 A flowchart of a robot kinematic parameter calibration method based on visual multi-point pose constraints is provided in an embodiment of the present invention, such as Figure 1 As shown, the calibration method includes:

[0051] Step 1: Control the robot to be calibrated with a camera installed at the end to move around the center of the calibration plate, and collect the robot model data in multiple posture states and the nominal posture data of the calibration plate during the movement.

[0052] Step 2: Use the hand-eye calibration algorithm to obtain the initial hand-eye conversion matrix as the initial value of the iteration.

[0053] Step 3: Generate a multi-point pose constraint equation group by combining the differential pose constraint equation groups of multiple calibration points on the calibration plate; the differential pose constraint equation group of a single calibration point is the relationship between the deviation between the nominal pose and actual pose of the calibration point and the deviation of the robot kinematic parameters.

[0054] Step 4: Use the weighted least squares method to iteratively solve the multi-point pose constraint equations to calibrate the kinematic parameters of the robot to be calibrated using the robot kinematic parameter deviation obtained when the iteration stops.

[0055] The present invention provides a robot kinematic parameter calibration method based on visual multi-point posture constraints, which can realize the identification and compensation of joint zero position error, connecting rod length error, and hand-eye conversion relationship, without measuring the actual position of the calibration plate during the process.

[0056] Example 1

[0057] Example 1 provided by the present invention is an embodiment of a robot kinematic parameter calibration method based on visual multi-point posture constraints provided by the present invention. The embodiment of the calibration method is based on a 6-axis robot closed-loop calibration method based on visual single-point constraints. A monocular camera is installed at the end of the 6R robot to form an eye-in-hand hand-eye system, and a machine vision calibration plate that can be recognized from multiple angles is placed at an appropriate position as a monocular posture measurement target. The robot hand-eye system is controlled to move around the center of the calibration plate, and data such as the robot joint angle in multiple posture states and the posture of the calibration plate in the camera coordinate system are collected to establish the 6R robot kinematic multi-point posture constraints. During the process, the actual position of the calibration plate does not need to be measured.

[0058] The camera is fixed to the robot's end effector to form an eye-in-hand system. The calibration plate is placed and fixed in an appropriate position as a measurement target, generally within the normal working range, so that the robot's eye-in-hand system can capture the measurement target from as many angles as possible. The actual position of the calibration plate does not need to be measured during the calibration process. The robot's eye-in-hand system is controlled to move around the calibration plate so that the calibration plate is as close to the center of the image as possible. When a single movement is completed, the camera image, visual pose measurement results, and robot joint angles are saved. Each calibration point is measured in multiple poses, and multiple sets of calibration point pose data are recorded.

[0059] Combine Figure 1 It can be seen that the embodiment of the calibration method includes:

[0060] Step 1: Control the robot to be calibrated with a camera installed at the end to move around the center of the calibration plate, and collect the robot model data in multiple posture states and the nominal posture data of the calibration plate during the movement.

[0061] In a possible embodiment, the robot model data collected in step 1 includes robot parameters and joint angles.

[0062] The nominal pose data of the calibration plate is obtained based on the visual pose measurement results of the robot.

[0063] The monocular vision pose measurement PnP (Perspective-n-Point) algorithm is used to measure pose and obtain the coordinate transformation relationship between the measurement target and the camera. Pose measurement requires the camera's intrinsic parameters. This paper adopts Zhang's calibration algorithm and uses the camera calibration toolbox to calibrate the camera's intrinsic parameters. The intrinsic parameter reprojection error is used as the accuracy indicator of the camera pose measurement.

[0064] The experiment provided by the embodiment of the present invention uses an asymmetric dot calibration plate as a measurement target. The visual measurement target (calibration plate) and the feature point extraction results are shown in the following example: Figure 2As shown in the figure, the MV-CA050-20UC camera and MVL-KF1628M-12MP fixed-focus lens are used, with a resolution of 2592x2048 and a field of view of approximately 20cm*20cm to maximize the pose accuracy of visual pose measurement.

[0065] The camera intrinsic parameters are calibrated using Zhang's method in OpenCV. The reprojection error is 0.15 pixels, and the visual pose measurement accuracy is about 0.015mm within a 20cm field of view.

[0066] Step 2: Use the hand-eye calibration algorithm to obtain the initial hand-eye conversion matrix as the initial value of the iteration.

[0067] Step 3: Generate a multi-point pose constraint equation group by combining the differential pose constraint equation groups of multiple calibration points on the calibration plate; the differential pose constraint equation group of a single calibration point is the relationship between the deviation between the nominal pose and actual pose of the calibration point and the deviation of the robot kinematic parameters.

[0068] In one possible embodiment, the process of obtaining the differential pose constraint equation group of a single calibration point includes:

[0069] Step 301: Under the pose constraint condition, the deviation between the nominal pose and the actual pose is calculated as the error model formula: ΔP = P Ab -P nb =f(w+Δw)-f(w).

[0070] Among them, ΔP represents the deviation; P Ab represents the actual posture, i P Ab =f(w+Δw) represents the actual position of point i on the calibration plate; P nb represents the nominal pose, i P nb =f(w) represents the nominal pose of the i-th point on the calibration plate; the subscript n represents the n-th measured pose of the i-th calibration point, w and Δw are the identifiable parameters of the robot model and their variations, respectively.

[0071] In step 302 , the error model formula is simplified by a first-order Taylor expansion and is abbreviated to: ΔP=JΔw, where J is the posture Jacobian matrix.

[0072] Step 303: Establish the i-th point differential pose constraint equation group as follows:

[0073] Abbreviated as i HΔw= i O.

[0074] Step 304: The multi-point pose constraint equations are generated by combining the single-point differential pose constraint equations and expressed as KΔw=B.

[0075] In a specific implementation, the position matrix and the Jacobian matrix are calculated based on the input data, and a multi-point pose constraint equation set KΔw=B is constructed, and the number of equations needs to be more than the number of identifiable parameters to avoid singular points and improve identification accuracy.

[0076] In a possible embodiment, the robot model is a DH (Denavit-Hartenberg) parameter model.

[0077] Identifiable parameters of the robot model:

[0078] w=[θ2,θ3,θ4,θ5,a2,a3,d3,R e0 ,R e1 ,R e2 ,R e3 ,T e0 ,T e1 ,T e2 ]。

[0079] Therefore, the change amount of the identifiable parameters is:

[0080] Δw=[Δθ2,Δθ3,Δθ4,Δθ5,Δa2,Δa3,Δd3,ΔR e0 ,ΔR e1 ,ΔR e2 ,ΔR e3 ,ΔT e0 ,ΔT e1 ,ΔT e2 ]

[0081] wherein θ2, θ3, θ4 and θ5 are correctable joint zero position parameters, a2, a3 and d3 are robot correctable link length parameters, [R e0 ,R e1 ,R e2 ,R e3 ] and [T e0 ,T e1 ,T e2 ] are hand-eye relationship parameters.

[0082] In a specific embodiment, the robot used in the present application is HSR-JR605 of Huashu, as shown in Figure 3 and Figure 4 are respectively a schematic diagram of a hand-eye system of the HSR-JR605 robot and a schematic diagram of a link coordinate system, combined with Figure 2 and Figure 3It can be seen that the first and second joint axes of the robot intersect at a point, and the fourth, fifth and sixth joint axes intersect at a point, and the nominal repeatability error is 0.02mm. Similar to other industrial robots, the absolute positioning accuracy is not mentioned in the product manual.

[0083] In order to realize fast and accurate kinematics inverse solution of the robot in actual industrial application, DH parameter method is used to establish kinematics model equation, and analytical inverse solution is used to complete inverse kinematics operation in motion control of the robot. The transformation relationship between the coordinate systems of the adjacent two links in the DH model can be described by four motion parameters, which are link torsion angle α i-1 , link length a i-1 , link offset d i , and joint angle θ i . As shown in Table 1, the link parameter index table of HSR-JR605 type industrial robot is shown.

[0084] Table 1 Link parameter of HSR-JR605 type industrial robot

[0085]

[0086] In the JR605 robot control system using the standard DH parameter model, d1 and θ1 can be converted into the pose parameters of the robot installation base, which can be calibrated after the robot is installed. The end effector is fixed with the camera, causing the joint 6 parameters to be coupled with the hand-eye parameters, so the parameter deviations of d6 and θ6 are included in the hand-eye parameters. Since the standard DH parameter model is used, the robot can correct the link length parameters a2, a3, d3, the joint zero position parameters θ2, θ3, θ4, θ5, and the hand-eye relationship parameters [R e0 , R e1 , R e2 , R e3 ] and [T e0 , T e1 , T e2 ], which are recorded as identifiable parameters w.

[0087] In one possible embodiment, the nominal pose of the calibration point on the calibration plate is

[0088] wherein, and are obtained from the transformation matrix of the adjacent joint link in the robot model ; wherein, i=1, 2, 3, 4, 5 or 6, base=0 and end=6 when the robot is a 6-DOF serial link robot; s represents the sine function; c represents the cosine function.

[0089] is the transformation matrix from the camera coordinate system to the robot end coordinate system, is the transformation matrix from the calibration plate coordinate system to the camera coordinate system.

[0090] In specific implementation, for a 6-DOF serial link robot, the transformation matrix of each adjacent link is multiplied to obtain the transformation matrix between the robot end and the robot base coordinate system:

[0091] This paper studies the robot kinematic calibration algorithm based on an eye-in-hand system in which a camera is mounted on the end of the robot. The coordinate system relationship of the eye-in-hand system is as follows:

[0092]

[0093] in, is the transformation matrix from the robot end coordinate system to the robot base coordinate system. When the DH parameters are fixed, Determined by the angles of each joint of the robot; is the transformation matrix from the camera coordinate system to the robot end coordinate system, which is an unknown quantity that needs to be calibrated. Since in the Eye-in-hand hand-eye system, the camera is fixed to the robot end, is an unknown constant; is the transformation matrix from the calibration plate coordinate system to the robot base coordinate system. The calibration plate is fixed during the calibration process. It is a constant unknown quantity.

[0094] To ensure the transformation matrix from the camera coordinate system to the robot end coordinate system To improve the stability of the rotation parameter identification, quaternions are used to represent the rotation parameters, so The hand-eye parameter R e and T e control:

[0095] R e =[R e0 ,R e1 ,R e2 ,R e3 ]

[0096] T e =[T e0 ,T e1 ,T e2 ]

[0097] Among them, the quaternion form is (w,x,y,z), R e0 2 +R e1 2 +R e2 2+R e3 2 =1.

[0098] Target pose under visual measurement conditions The positive solution equation is:

[0099] Should That is the nominal pose f(w) of the calibration point on the calibration plate.

[0100] In a possible embodiment, in step 302, the error model formula is simplified by a first-order Taylor expansion to:

[0101]

[0102] Since the pose matrix T is a homogeneous transformation matrix, it can be expressed as where r x 、r y 、r z Follow the right-hand rule x ×r y =r z Due to the coupling of the rotation matrix, we take r x 、r y , p is simplified as the pose eigenvalue, so J is 9*14 and is the pose Jacobian matrix.

[0103] In one possible embodiment, the process of establishing the single-point differential pose constraint equation group in step 303 is:

[0104] Based on the nominal pose data of the calibration plate obtained by visual pose measurement at different poses of the robot in the workspace, a single-point pose constraint equation group is established:

[0105]

[0106] Eliminate the actual pose of point i in the pose constraint equations by using the difference method i P Ab , and obtain the single-point difference pose constraint equations.

[0107] Therefore, this calibration method does not require measuring the actual position of the calibration plate i P Ab .

[0108] Step 4: Use the weighted least squares method to iteratively solve the multi-point pose constraint equations to calibrate the kinematic parameters of the robot to be calibrated using the robot kinematic parameter deviation obtained when the iteration stops.

[0109] Since the multi-point pose constraint equations KΔw=B are overdetermined, this paper uses the weighted least squares method for iterative solution. In a possible embodiment, using the weighted least squares method in step 4 to solve the multi-point pose constraint equations KΔw=B is equivalent to solving (K T K+λI)Δw=K T B equation; where λ is the weight, and the weight λ is passed through B T The changing trend of B is optimized with variable step size.

[0110] In a possible embodiment, the condition for stopping iteration in step 4 includes: satisfying a maximum number of iterations or an end winding point error, etc.

[0111] The calculation process of the terminal winding error includes: measuring m points in multiple postures for n calibration points n The average of the position errors of each pose relative to the point is taken as the terminal point error.

[0112] The specific formula is:

[0113] To improve the kinematic absolute positioning accuracy of a six-axis robot, this paper provides a six-axis robot kinematic closed-loop calibration method based on visual pose measurement technology. First, based on the inverse parameter constraints and hand-eye system kinematic coupling relationship in industrial applications, the minimum parameter set for kinematic calibration in industrial environments is extracted. A pose differential method is used based on high-precision visual pose results to establish a kinematic closed-loop constraint optimization equation. Parameter optimization is achieved using a weighted least squares method, without measuring the actual position of the calibration plate. This method can identify and compensate for joint zero-position errors, link length errors, and hand-eye conversion relationships. Results show that the average distance the robot's end-point orbits a point is reduced from 1.62 mm to 0.27 mm.

[0114] Example 2

[0115] Example 2 provided by the present invention is an application example of a robot kinematic parameter calibration method based on visual multi-point pose constraints provided by the present invention. To verify the effectiveness of the calibration method provided by the present invention, a robot experimental platform was built, using the 6R robot Wasu HSR-JR605 as the target robot, and a monocular camera was installed at the end to form an eye-in-hand hand-eye system. Figure 5 This diagram illustrates an experimental system for an application example of the present invention's method for calibrating robot kinematic parameters based on multi-point visual pose constraints. A machine vision calibration plate, capable of multi-angle recognition, is placed at an appropriate location as a monocular pose measurement target. The robot is controlled to move around the center of the plate, collecting data such as the robot's joint angles in multiple poses and the plate's pose in the camera coordinate system.

[0116] In the calibration experiment of the JR605 robot, the measurement data included 64 sets of multi-pose measurement data of 3 calibration points. The calibration experiment was carried out using the algorithm proposed in this paper. After 200 iterations, the calibration error converged to 0.27mm. The parameter identification results are shown in Table 2. After calibration, the average distance error of the robot end point around the point was reduced from 1.62mm to 0.27mm. Figure 6 As shown, the error distribution histogram before and after calibration is as follows Figure 7 As shown, it can be seen that the error is smaller and the error distribution is more concentrated after calibration.

[0117] Table 2 Compensation values ​​of JR605 kinematic parameters after calibration

[0118]

[0119]

[0120] Conversion parameters from camera coordinate system to end coordinate system:

[0121] T=[64.125,-26.971,258.558]

[0122] R=[0.7196,-0.0464,0.0390,0.6917]

[0123] An embodiment of the present invention provides a multi-point posture constraint robot closed-loop kinematic calibration method based on a visual posture measurement method. Based on the monocular visual posture measurement method, the robot hand-eye system end is controlled to shoot and measure the fixed calibration plate target in multiple postures to realize the posture measurement in the robot kinematic calibration. Multi-point posture constraints are established based on the kinematic equations of the robot hand-eye system, and the posture difference method is used based on the high-precision visual posture measurement results to establish the kinematic closed-loop constraint optimization equation, and the weighted least squares method is used to realize parameter optimization. The relative posture measurement of the end is realized based on the monocular visual posture measurement method, and the automation level of the measurement process is improved under the condition of higher measurement accuracy. Beneficial effects include:

[0124] First, the camera in a common robot hand-eye system can be directly utilized at a low cost. Kinematic calibration can be performed in conjunction with a low-cost calibration plate. The equipment requires low installation precision, making it easy to install, disassemble, and carry. The calibration process does not require measuring the actual position of the calibration plate. Second, the high-precision calibration plate target provides the foundation for high-precision measurement. The sub-pixel feature extraction algorithm for the calibration plate target and the monocular vision pose method further improve the system's measurement accuracy and stability. The system calibration process is then performed at the robot workstation, eliminating the need for robot disassembly or transportation. There are no special workspace requirements for calibration. Furthermore, the low cost and easy installation of the calibration tool enable long-term, on-site accuracy monitoring.

[0125] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0126] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0130] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims include all such modifications and variations as fall within the scope of the present application.

[0131] It is apparent that those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. It is therefore intended that the present application cover all such changes and modifications that are within its scope.

Claims

1. A robot kinematic parameter calibration method based on visual multi-point pose constraints, characterized in that: The calibration method comprises: Step 1: Control the robot to be calibrated with a camera installed at the end to move around the center of the calibration plate, and collect robot model data in multiple posture states during the movement and the nominal posture data of the calibration plate; Step 2: Use the hand-eye calibration algorithm to obtain the initial hand-eye conversion matrix as the initial value of the iteration; Step 3, generating a multi-point pose constraint equation group by combining the differential pose constraint equation groups of multiple calibration points on the calibration plate; the differential pose constraint equation group of a single calibration point is the relationship between the deviation between the nominal pose and the actual pose of the calibration point and the deviation of the robot kinematic parameters; Step 4, using a weighted least squares method to iteratively solve the multi-point pose constraint equations, so as to calibrate the kinematic parameters of the robot to be calibrated by using the robot kinematic parameter deviation obtained when the iteration stops; The process of obtaining the differential pose constraint equations of a single calibration point includes: Step 301: Calculate the deviation between the nominal pose and the actual pose as the error model formula: ; in, express deviation; represents the actual posture, Indicates the actual pose of point i on the calibration plate; represents the nominal pose, Indicates the nominal pose of the i-th point on the calibration plate; subscript n indicates the n-th measured pose of the i-th calibration point. and are the identifiable parameters of the robot model and their variations; Step 302: Simplify the error model formula using a first-order Taylor expansion to: ; is the pose Jacobian matrix; Step 303: Establish a single-point differential pose constraint equation set as follows: , abbreviated as ; Step 304: construct a multi-point pose constraint equation group by combining the single-point differential pose constraint equation groups to form a multi-point pose constraint equation group represented as , abbreviated as ; The robot model is a DH parameter model; Identifiable parameters of the robot model: ； in, 、 、 and To correct the joint zero position parameters, 、 、 is the robot's adjustable link length parameter, and is the hand-eye relationship parameter.

2. The calibration method according to claim 1, wherein: The robot model data collected in step 1 includes robot parameters and joint angles; The nominal pose data of the calibration plate is obtained based on the visual pose measurement result of the robot.

3. The calibration method according to claim 1, wherein: Nominal pose of the calibration points on the calibration plate ; in, 、 、 、 、 and The transformation matrix of adjacent joint links in the robot model Get; wherein, i=1, 2, 3, 4, 5 or 6, when the robot is a 6-degree-of-freedom serial link robot, base=0, end=6; s represents a sine function; c represents a cosine function; is the transformation matrix from the camera coordinate system to the robot end coordinate system, is the transformation matrix from the calibration plate coordinate system to the camera coordinate system.

4. The calibration method according to claim 1, wherein: In step 302, the error model formula is simplified by a first-order Taylor expansion to: 。 5. The calibration method according to claim 1, wherein: The process of establishing the single-point differential pose constraint equation group in step 303 is as follows: Based on the nominal pose data of the calibration plate obtained by visual pose measurement at different robot poses, a single-point pose constraint equation group is established: ； Eliminate the actual pose of point i in the pose constraint equations by using the difference method , the single-point differential pose constraint equation group is obtained; the multi-point pose constraint equation group is generated by combining the single-point differential pose constraint equation groups and expressed as , abbreviated as .

6. The calibration method according to claim 1, wherein: In step 4, the weighted least squares method is used to solve the multi-point pose constraint equations Equivalent to solving Equation; where is weight, weight pass The variable step size optimization is performed based on the changing trend of .

7. The calibration method according to claim 1, wherein: The conditions for stopping iteration in step 4 include: meeting the maximum number of iterations or the terminal winding point error; The calculation process of the terminal winding point error includes: multi-posture measurement of n calibration points The average of the position errors of each pose relative to the point is taken as the terminal point error.

Images