Industrial robot tcp calibration device and calibration method

Abstract

The application discloses an industrial robot TCP calibration device and a calibration method. The calibration method respectively allocates two different robot end tool postures, translates the robot to make the TCP not collinear and touch the corresponding plane at least three times; singular value decomposition is performed on the flange position point set collected for each posture, and a direction vector corresponding to the minimum singular value is extracted as the normal estimation value of the touched plane corresponding to the posture; for each plane, a linear equation containing only the unknown TCP is constructed according to the flange position centroid difference and the flange posture rotation matrix difference of the two postures, and the three-dimensional bias of the TCP in the flange coordinate system is solved. When the calibration method of the method is calibrated, the robot only needs to perform translational motion, does not need to perform posture transformation, and does not need to accurately control the touch position, so that the operation difficulty is greatly reduced, the full-automatic calibration process is easy to realize, the robustness is high, and the calibration quality is completely quantifiable.

Description

Technical Field

[0001] This invention belongs to the field of industrial robot calibration technology, specifically relating to a device and method for automatically calibrating the tool center point (TCP) of a robot end effector using a triorthogonal plane calibration block and dual-attitude contact measurement, which is particularly suitable for rapid on-site calibration of the TCP of a welding robot welding torch. Background Technology

[0002] In industrial robot applications such as welding, cutting, and grinding, there is a three-dimensional offset between the tool center point of the end effector (such as a welding torch) and the robot flange coordinate system. Accurately calibrating this TCP offset is a prerequisite for ensuring the positioning accuracy of the robot end effector.

[0003] Existing TCP labeling methods can be mainly divided into the following categories: (1) Four-point method / multi-point method: The TCP touches the same fixed cusp in space with different postures, and the TCP is solved by establishing a system of equations using position invariant constraints. This method requires cusp calibration tools and requires the TCP to touch the same spatial point accurately with at least 4 different postures. It has high technical requirements for operators, and the sliding and contact detection delay when touching the cusp can easily introduce errors.

[0004] (2) Plane method: The TCP touches a plane and the TCP is solved using the plane constraint equation. The single plane method only provides one constraint equation and requires at least three different planes to determine the three-dimensional coordinates of the TCP; however, the relative positional relationship (normal and distance) between the planes is unknown, resulting in too many unknowns and an underdetermined system. It usually requires the introduction of prior knowledge of the geometric relationship between the planes or iterative optimization, which makes the solution complex.

[0005] (3) Laser tracker method: The TCP position is directly measured using an external laser tracker. It has high accuracy but the equipment is expensive and is not suitable for rapid on-site calibration.

[0006] (4) Visual method: Use a camera to identify and calibrate the target and calculate the TCP position. This requires complex image processing and is sensitive to ambient lighting.

[0007] In summary, existing methods struggle to simultaneously meet the demands for low cost, high precision, full automation, and rapid on-site TCP calibration. In particular, achieving fully automated TCP calibration with closed-form analytical solutions without requiring precise apex alignment, expensive laser trackers, or iterative optimization of initial values ​​is a pressing issue in this field. Summary of the Invention

[0008] The technical problem to be solved by this invention is to overcome the defects of the prior art and provide an industrial robot TCP calibration device and method. This device and method only requires three orthogonal plane calibration blocks and two different posture touches on each face to achieve closed-loop analytical solution of TCP three-dimensional coordinates, thus resolving at least one of the following technical contradictions in existing TCP calibration methods: (1) The cusp method requires precise contact with the same point, which is difficult to operate; (2) The number of unknowns in the single-plane method is greater than the number of equations, requiring iteration or prior knowledge; (3) Laser method equipment is expensive, and visual method is affected by environmental interference.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: An industrial robot TCP calibration device includes three mutually orthogonal metal planes that intersect at a corner point to form an interior cube structure. During the calibration process, the calibration device is fixed in the robot's workspace, and the position and orientation of the calibration device do not need to be pre-calibrated.

[0010] Furthermore, each of the planes is made of conductive metal, with a flatness error ≤0.05 mm and an orthogonality error between adjacent planes ≤0.05°.

[0011] A calibration method based on an industrial robot TCP calibration device includes the following steps: S1: Assign two different robot end-effector postures to each of the three planes of the calibration device. Translate the robot so that the TCP does not collinearly touch the corresponding plane at least three times. Record the robot flange's pose in the base coordinate system each time the contact is made. S2: Perform singular value decomposition on the flange position point set acquired for each attitude, and extract the direction vector corresponding to the minimum singular value as the normal estimate of the plane touched by that attitude; after aligning the directions of the two normal estimates of the same plane, take the vector sum and normalize to obtain the merged normal of the plane. S3: Perform singular value decomposition on the direction matrix formed by the combined normals of the three planes, and take the product of the left singular matrix and the right singular matrix as the nearest orthogonal matrix. Its column vectors are the unit normals of the three planes after orthogonalization correction. S4: For each plane, using the orthogonally corrected unit normal, the difference in the centroid of the flange position between the two orientations of the plane, and the difference in the flange orientation rotation matrix, construct a linear equation containing only TCP unknowns; the three planes form a 3×3 system of linear equations, and solve it to obtain the three-dimensional offset of TCP in the flange coordinate system.

[0012] Further, in step S2, the specific steps for calculating the normal estimate of the plane are as follows: calculate the centroid of the flange position point set for each attitude and construct a centered position matrix; perform singular value decomposition on the centered position matrix and extract the left singular vector corresponding to the minimum singular value as the normal estimate of the plane touched by that attitude.

[0013] Furthermore, in step S3, the orthogonalization correction is to compensate for the normal estimation error and the calibration device manufacturing error: by solving for the orthogonal matrix that is closest to the estimated direction matrix, the three normals are corrected to be strictly orthogonal.

[0014] Further, in step S4, the process of constructing the linear equation is as follows: let the orthogonally corrected unit normal of plane k be ñ_k, the directed distance from the plane to the origin of the base coordinate system be d_k, and the TCP bias vector be t; The constraint equations for the two orientations k1 and k2 of the plane are ñ_k·P_bar_k1 + ñ_k·R_k1·t = d_k and ñ_k·P_bar_k2 + ñ_k·R_k2·t = d_k. Subtracting d_k from the two equations, we get ñ_k·(R_k1 - R_k2)·t = -ñ_k·(P_bar_k1 - P_bar_k2). In the formula, P_bar_k1 is the centroid of the flange position vector of all contact points in attitude k1, P_bar_k2 is the centroid of the flange position vector of all contact points in attitude k2, R_k1 is the rotation matrix of the flange coordinate system under attitude k1 relative to the base coordinate system, and R_k2 is the rotation matrix of the flange coordinate system under attitude k2 relative to the base coordinate system.

[0015] Furthermore, it also includes the S5 quality assessment step, which evaluates the reliability of the attitude normal estimation, the risk of collinearity, the difference between two attitudes on the same plane, the overall calibration accuracy, and / or cross-attitude consistency.

[0016] Furthermore, the reliability of the attitude normal estimation is evaluated by the ratio of singular values ​​in the SVD decomposition, i.e., the singular value of the normal / the singular value of the orthogonal direction in the plane. The risk of collinearity at sampling points is evaluated by the singular values ​​of the orthogonal directions in the plane / the singular values ​​of the main extension direction. The sufficiency of the difference in attitude angles between two objects on the same plane is evaluated by the condition number of the 3×3 linear system matrix. The overall calibration accuracy is evaluated by the root mean square of the residuals of the equations at each touch point. Cross-pose consistency is assessed by independently estimating the directed distance between two poses on the same plane.

[0017] Furthermore, the rotation matrix of the two end-effector postures on the same plane satisfies ||R_k1 - R_k2|| ≥ 0.005, and the angle between the Z-axis direction of the end-effector and the plane normal is within the range of 0° to 60°, in order to accommodate welding gun tools with protective sleeves.

[0018] Furthermore, in step S1, the number of sampling points for each attitude is no less than 3 and they are not collinear; the total number of sampling points is no less than 18; when the number of sampling points for a certain attitude exceeds 3, the influence of measurement noise on normal estimation is reduced by SVD least squares fitting; when the number of sampling points for a certain attitude is equal to 3, the plane determined by the three points directly gives the plane normal.

[0019] The principle behind this solution: Let n_k be the unit normal of plane k (k = A, B, C), d_k be the directed distance from the plane to the origin of the base coordinate system, and t ∈ R³ be the TCP bias.

[0020] For the poses k1 and k2 on plane k, the flange poses (R_k1, P_k1,i) and (R_k2, P_k2,j) are acquired respectively. The constraint equations at the time of contact are: n_k · P_k1_i + n_k · R_k1 · t = d_k(1) n_k · P_k2_j + n_k · R_k2 · t = d_k(2).

[0021] In the above formula, R_k1 and R_k2 are the 3×3 rotation matrices of the flange coordinate system {F} relative to the base coordinate system {B} under two attitudes of plane k; P_k1_i and P_k2_j are the 3×1 position vectors of the origin of the flange coordinate system in the base coordinate system at the i / jth contact in the corresponding attitude.

[0022] Key Step 1 (Independence of Normal Estimation): Under the same posture, the TCP position difference between two touches is equal to the flange position difference: X_i - X_j = (P_k1-i + R_k1·t) - (P_k1-j + R_k1·t) = P_k1-i - P_k1-j.

[0023] In the formula, X_i and X_j represent the three-dimensional position coordinates of TCP in the robot's base coordinate system at the i-th and j-th touches, respectively.

[0024] The TCP difference vector lies in the plane, and the flange position difference vector also lies in the plane. Therefore, the plane normal can be obtained by performing SVD decomposition on the flange position point set without knowing the TCP: the direction with the minimum variance of the flange position point set is the plane normal.

[0025] Key Step Two (Planar Distance Elimination): Substitute the centroids of the flange positions in equations (1) and (2) and subtract them: n_k·(R_k1 - R_k2)·t = -n_k·(P_bar_k1 - P_bar_k2). The planar distance d_k in the equation is completely eliminated, containing only the unknown t. Each of the three planes provides an equation, forming a 3×3 linear system F·t = b.

[0026] Since the normals of the three planes are mutually orthogonal, and there is a sufficient angular difference between the two orientations of each plane (R_k1 ≠ R_k2), the matrix is ​​non-singular, t = F - ¹·b has a unique solution.

[0027] Compared with the prior art, the present invention has the following beneficial effects: (1) Fully closed analytical solution: No iterative optimization is required, it does not depend on initial values, and the calculation is stable and reliable. The entire calibration process consists of three SVD decompositions and one solution of a 3×3 linear equation system. The overall computational load increases linearly with the number of data points involved in the calibration.

[0028] (2) The calibration device is simple and low-cost: it only requires three mutually orthogonal conductive planes, without the need for precision tips, active marking or targets, and the manufacturing and maintenance costs are extremely low.

[0029] (3) Simple and fully automatic operation: Each side uses a fixed tool posture for contact. The robot only needs to perform translational movement without posture change. The contact position does not need to be precisely controlled (it only needs to be distributed on the plane), which greatly reduces the difficulty of operation and makes it easy to realize a fully automatic calibration process.

[0030] (4) Strong robustness: The plane normal is fitted by the difference of the robot flange position, which is independent of the TCP true value; the plane distance is eliminated by the difference of the centroids of the two attitudes, and the original underdetermined problem of 3 equations and 6 unknowns is transformed into a well-determined problem of 3 equations and 3 unknowns; the redundant sampling points are naturally denoised by centroid averaging.

[0031] (5) Quantifiable quality assessment: The reliability of the normal estimate is evaluated by the SVD singular value ratio, the sufficiency of attitude difference is evaluated by the 3×3 matrix condition number, and the overall calibration consistency is evaluated by the point-by-point residual. The calibration quality is fully quantifiable.

[0032] (6) Relaxed attitude constraints: It only requires that there is an angle difference between the two attitudes on the same surface (the greater the difference, the better the condition number). There are no strict constraints on the angle relationship between the tool axis and the plane normal (it can be within the range of 0°~60°), which is suitable for restricted tools such as welding guns with protective sleeves. Attached Figure Description

[0033] Figure 1Schematic diagram of the coordinate system of the calibration device.

[0034] Figure 2 Calibration method flowchart.

[0035] Figure 3 : A schematic diagram of the geometric principle of SVD normal estimation in step S2.

[0036] Figure 4 : A schematic diagram illustrating the principle of eliminating planar distance by the difference in centroids between two attitudes in step S4.

[0037] Figure 5 : A schematic diagram showing the correspondence between the 6 tool postures and the 3 planes, and the distribution of sampling points. Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0039] Example 1: This embodiment discloses an industrial robot TCP calibration device, such as... Figure 1 As shown, the system comprises three mutually orthogonal metal planes that intersect at a single corner, forming an interior cube structure. Each plane is made of a conductive metal (such as stainless steel or aluminum alloy), with a flatness error ≤0.05 mm and an orthogonality error between adjacent planes ≤0.05°. The calibration device is fixed within the robot's workspace during the calibration process; its position and orientation do not need to be precisely known.

[0040] The method for automatic TCP calibration of end-effector tools (such as welding torches) using the above-mentioned calibration device is as follows: Figures 2 to 5 As shown, it includes the following steps: Step S1: Data Acquisition Two different robot tool poses (a total of 6 poses) are assigned to the three planes of the calibration device. The TCP of each pose touches its corresponding plane at least 3 times (not collinear). The pose of the robot flange in the base coordinate system is recorded each time the contact is activated.

[0041] For ease of understanding, the main mathematical symbols involved in this method are uniformly defined in the following table:

[0042] The specific steps of this calibration method are as follows: S1.1 Fix the tri-orthogonal plane calibration device within the robot's workspace. The intersection point of the three planes of the calibration device is located within the robot's reachable workspace; the exact location does not need to be known precisely.

[0043] S1.2 Set 6 tool postures, labeled A1, A2 (corresponding to plane A), B1, B2 (corresponding to plane B), C1, C2 (corresponding to plane C), as follows: Figure 5 As shown.

[0044] Let n_k be the unit normal of plane k (k∈A,B,C), d_k be the directed distance from plane k to the origin of the robot's base coordinate system, and t∈R³ be the bias of TCP, where R is a 3×1 real column vector.

[0045] For poses k1 and k2 on plane k, the robot flange poses (R_k1, P_k1_i) and (R_k2, P_k2_j) are acquired respectively. The constraint equations at the moment of contact are then: n_k P_k1_i+n_k R_k1 t=d_k(1) n_k P_k2_j+n_k R_k2 t=d_k(2) For two orientations of the tool on the same plane, the Z-axis directions are different, and the rotation matrix satisfies ||R_k1 - R_k2||>0.005.

[0046] In the formula, ||·|| represents the Frobenius norm of the matrix (i.e., the square root of the sum of the squares of all elements), used to measure the degree of difference between two rotation matrices. A threshold of 0.005 corresponds to an attitude difference angle of approximately 1.6°.

[0047] S1.3 The welding wire at the tip of the welding torch is energized. For each posture, the tool posture remains unchanged, and only translational movement is performed; the tip of the welding wire is slowly brought close to the corresponding plane, and the acquisition is triggered at the moment of conduction to record the flange posture; the contact is repeated at least 3 times, and the contact points are distributed in a dispersed manner on the plane (not collinear); the coverage area of ​​the contact points is preferably greater than 20×20mm to obtain a stable normal estimate.

[0048] S1.4 Record data format: Record the pose of the robot flange in the base coordinate system each time the circuit is activated. Each record contains an attitude identifier, a 3×3 rotation matrix, and a 3×1 position vector.

[0049] Step S2: Normal estimation, such as Figure 3 As shown, for each tool posture, the N flange position point set is used to extract the direction vector corresponding to the minimum singular value using singular value decomposition (SVD) as the normal estimate of the plane touched by that posture; the two normal estimates of the same plane are averaged after direction alignment to obtain the combined normal of the plane.

[0050] The specific steps are as follows: S2.1 For each flange position point set of attitude m, calculate the centroid P_bar, denoted as P_bar. , is the centroid (arithmetic mean) of the flange position vectors of all contact points in this posture.

[0051] Construct a centralized position matrix M (3×N), where the i-th column of the centralized matrix M is P_i. That is, the coordinates of each flange location point minus the centroid.

[0052] Perform SVD decomposition on M: M = U·Sigma·V U is a 3×3 left singular matrix, whose column vectors correspond to the three orthogonal principal directions of the flange position point set; Sigma is a 3×N diagonal singular value matrix, where the diagonal elements are singular values ​​sigma_1≥sigma_2≥sigma_3≥0; V Let V be the transpose of an N×N right singular matrix, and T denote the transpose operation.

[0053] The left singular vector corresponding to the minimum singular value is taken as the estimated normal vector of the plane touched by this attitude, n_hat=U[:,2], where U[:,2] represents the direction with the minimum variance of the flange position point set, which is the 3rd column of matrix U (column index starts from 0 and corresponds to the minimum singular value σ3), and is taken as the estimated normal vector of the plane. This estimated normal vector is independent of the unknown TCP value. To ensure consistency of direction, the sign of the component with the largest normal magnitude is taken as positive.

[0054] Key step -- Independence of normal estimation: Under the same attitude, the TCP position difference between two touches is equal to the flange position difference: X_i-X_j=(P_k1_i+R_k1·t)-(P_k1_j+R_k1·t)=P_k1_i-P_k1_j.

[0055] The TCP difference vector lies in the plane, and the flange position difference vector also lies in the plane. Therefore, the plane normal can be obtained by performing SVD decomposition on the flange position point set without knowing the TCP: the direction with the minimum variance of the flange position point set is the plane normal.

[0056] S2.2 Merging two normals of the same plane: If the dot product of two normal estimates is negative, invert one of them. After aligning the directions, take the average to obtain the merged normal of the plane, which is a vector sum normalized.

[0057] S2.3 Calculate the coplanar normal consistency index θ_k=arccos(|n_hat_k1) If n_hat_k2|), a warning is issued if the angle is greater than 2°, indicating that the quality of the two attitude points on the same surface is inconsistent.

[0058] Step S3: Orthogonalization Singular value decomposition is performed on the direction matrix formed by the combined normals of the three planes. The product of the left and right singular matrices is used as the nearest orthogonal matrix to obtain three mutually orthogonal unit normals.

[0059] The specific steps are as follows: S3.1 Construct the direction matrix N_hat=[n_hat_A,n_hat_B,n_hat_C] (3×3).

[0060] S3.2 Perform SVD decomposition on N_hat: N_hat = U_N·Sigma_N·V_N .

[0061] In the formula, the subscript N indicates that the matrix is ​​derived from the SVD decomposition of the normal matrix N_hat, to distinguish it from the SVD decomposition (U, Sigma, V) of the position point set M in step S2. Both U_N and V_N are 3×3 orthogonal matrices.

[0062] S3.3 Nearest Orthogonal Matrix N = U_N·V_N If det(N) = -1, invert the third column.

[0063] S3.4 The columns with normal N after orthogonalization. Calculate the orthogonalization correction angle δ_k = arccos(n_hat_k·ñ_k). The orthogonalization correction angle reflects the magnitude of the correction and serves as a verification indicator for the manufacturing accuracy of the calibration device.

[0064] Orthogonalization correction is a compensation for the normal estimation error and the manufacturing error of the calibration device: by solving for the orthogonal matrix that is closest to the estimated direction matrix in the sense of Frobenius norm, the three normals are corrected to be strictly orthogonal.

[0065] Step S4: TCP solution, such as Figure 4 As shown, for each plane, a linear equation containing only TCP unknowns is constructed using the difference in centroids of the flange positions and the difference in attitude rotation between the two attitudes of that plane; the three planes form a 3×3 system of linear equations, which can be directly solved to obtain the TCP three-dimensional offset.

[0066] The specific steps are as follows: S4.1 For each plane k, calculate the difference in flange position centroid b_k and the difference in attitude rotation a_k between the two attitudes of the plane: a_k=ñ_k^T·(R_k1-R_k2), b_k=-ñ_k^T·(P_bar_k1-P_bar_k2).

[0067] In the formula, ñ_k is the unit normal vector (3×1 column vector) after orthogonalization correction in step S3, and its superscript T indicates transpose, so ñ_k^T is a 1×3 row vector; a_k is a 1×3 row vector, and b_k is a scalar. The a_k and b_k of the three planes constitute the elements of the row vectors and 3×1 column vector b of the 3×3 matrix F, respectively.

[0068] S4.2 Assemble a 3×3 linear system F·t=b, where the row vectors of F are a_A,a_B,a_C; b=[b_A,b_B,b_C]^T.

[0069] S4.3 Solve for t=F -1 b. If the condition number cond(F) of matrix F > 100, a warning is issued, indicating insufficient difference in pose on the same surface.

[0070] Key step – Planar distance elimination: Substitute the centroids of the flange positions in equations (1) and (2) and subtract them. Simultaneously, the unit normal of plane k is replaced by the orthogonally corrected unit normal ñ_k. Then: ñ_k·(R_k1-R_k2)·t=-ñ_k·(P_bar_k1-P_bar_k2). The planar distance d_k in the equations is completely eliminated, containing only the unknown t. Each of the three planes provides an equation, forming a 3×3 linear system F·t=b.

[0071] Since the normals of the three planes are mutually orthogonal, and there is a sufficient angular difference between the two orientations of each plane (R_k1≠R_k2), the matrix is ​​non-singular, t=F -1 b has a unique solution.

[0072] S4.4 Optional step: Calculate the planar distance d_k and corner coordinates using TCP back-substitution for quality verification.

[0073] Step S5: Quality Assessment (1) SVD coplanarity: The minimum singular value sigma_3 of SVD for each attitude reflects the out-of-plane dispersion of the flange sampling point and should be close to the measurement noise level.

[0074] sigma_1, sigma_2, and sigma_3 are the three singular values ​​obtained from the SVD decomposition of each attitude position point set in step S2, arranged in descending order: sigma_1 ≥ sigma_2 ≥ sigma_3 ≥ 0. sigma_1 is the singular value of the principal extension direction (the direction of maximum dispersion in the plane), sigma_2 is the singular value of the orthogonal direction in the plane, and sigma_3 is the singular value of the out-of-plane direction (normal). The reliability of the normal estimation for a single attitude is evaluated by the ratio of the singular values ​​from the SVD decomposition, sigma_3 / sigma_2. When the ratio > 0.3, it indicates that the out-of-plane fluctuation of the sampling point is large or the in-plane dispersion is insufficient, and the normal estimation may be unreliable.

[0075] (2) Collinearity check: sigma_2 / sigma_1<0.01 indicates that the sampling points are close to collinear and the normal estimate is unreliable.

[0076] (3) Condition number check: cond(F)<50 is good, 50~100 is average, and >100 requires adjustment of the same-plane posture difference.

[0077] (4) Point-by-point residual: r_m_i=ñ_k·P_m_i+ñ_k·R_m·t-d_k, whose RMS value reflects the overall calibration accuracy.

[0078] In the formula, m is the attitude identifier (m ∈ {A1, A2, B1, B2, C1, C2}), i is the contact point number in attitude m; ñ_k is the unit normal of the plane k corresponding to the attitude after orthogonalization correction; d_k is the directed distance from the plane to the origin of the base coordinate system calculated by substituting the calibration result t back into formula (1) or (2).

[0079] (5) Cross-attitude consistency: The difference between the directed distance d_k estimated independently for two attitudes on the same plane should be close to zero.

[0080] Example 2: Based on Embodiment 1, the technical solution of Embodiment 1 can be improved. The following are several improved alternative implementation methods: Option 1: Variable Number of Points Implementation: In step S2, the SVD normal fitting accepts any number of points (≥3), and in step S4, the centroid calculation is automatically adapted. When only 3 points are collected for a certain pose, the normal is uniquely determined by the plane defined by these 3 points; when more points are collected, the noise impact is reduced through SVD least squares fitting. Other aspects are the same as in Example 1.

[0081] Option 2: The contact detection method in step S1 can be any of the following: (a) Conductive contact type - the welding wire is energized and conducts through the contact metal plane (suitable for conductive tools and metal calibration devices). (b) Force sensor type - A force sensor is installed at the end of the tool to detect the contact force threshold and trigger the acquisition (suitable for non-conductive tools). (c) Acoustic emission type – the acquisition is triggered by the acoustic emission signal at the moment of contact. The rest is the same as the scheme in Example 1.

[0082] Option 3: Automated Process: Step S1 can be executed automatically by the robot—the approximation direction and search step size for each posture are preset, and the conduction signal automatically triggers pose recording and retracement. The entire process requires no manual intervention. Other aspects are the same as in Example 1.

[0083] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An industrial robot TCP calibration device, characterized in that, It includes three mutually orthogonal metal planes, which intersect at a corner point to form an interior cube structure; during the calibration process, the calibration device is fixed in the robot's workspace, and the position and orientation of the calibration device do not need to be calibrated in advance.

2. The industrial robot TCP calibration device according to claim 1, characterized in that, Each of the planes is made of conductive metal, with a flatness error of ≤0.05 mm and an orthogonality error between adjacent planes of ≤0.05°.

3. A calibration method based on the industrial robot TCP calibration device of claim 1, characterized in that, Includes the following steps: S1: Assign two different robot end-effector postures to each of the three planes of the calibration device. Translate the robot so that the TCP does not collinearly touch the corresponding plane at least three times. Record the robot flange's pose in the base coordinate system each time the contact is made. S2: Perform singular value decomposition on the flange position point set acquired for each attitude, and extract the direction vector corresponding to the minimum singular value as the normal estimate of the plane touched by that attitude; after aligning the directions of the two normal estimates of the same plane, take the vector sum and normalize to obtain the merged normal of the plane. S3: Perform singular value decomposition on the direction matrix formed by the combined normals of the three planes, and take the product of the left singular matrix and the right singular matrix as the nearest orthogonal matrix. Its column vectors are the unit normals of the three planes after orthogonalization correction. S4: For each plane, using the orthogonally corrected unit normal, the difference in the centroid of the flange position between the two orientations of the plane, and the difference in the flange orientation rotation matrix, construct a linear equation containing only TCP unknowns; the three planes form a 3×3 system of linear equations, and solve it to obtain the three-dimensional offset of TCP in the flange coordinate system.

4. The industrial robot TCP calibration method according to claim 3, characterized in that, In step S2, the specific steps for calculating the normal estimate of the plane are as follows: calculate the centroid of the flange position point set for each attitude and construct a centered position matrix; perform singular value decomposition on the centered position matrix and extract the left singular vector corresponding to the minimum singular value as the normal estimate of the plane touched by that attitude.

5. The industrial robot TCP calibration method according to claim 3, characterized in that, In step S3, the orthogonalization correction is to compensate for the normal estimation error and the calibration device manufacturing error: by solving for the orthogonal matrix that is closest to the estimated direction matrix, the three normals are corrected to be strictly orthogonal.

6. The industrial robot TCP calibration method according to claim 3, characterized in that, In step S4, the process of constructing the linear equation is as follows: Let the orthogonally corrected unit normal of plane k be ñ_k, the directed distance from the plane to the origin of the base coordinate system be d_k, and the TCP bias vector be t; The constraint equations for the two orientations k1 and k2 of the plane are ñ_k·P_bar_k1 + ñ_k·R_k1·t = d_k and ñ_k·P_bar_k2 + ñ_k·R_k2·t = d_k. Subtracting d_k from the two equations, we get ñ_k·(R_k1 - R_k2)·t = -ñ_k·(P_bar_k1 - P_bar_k2). In the formula, P_bar_k1 is the centroid of the flange position vector of all contact points in attitude k1, P_bar_k2 is the centroid of the flange position vector of all contact points in attitude k2, R_k1 is the rotation matrix of the flange coordinate system under attitude k1 relative to the base coordinate system, and R_k2 is the rotation matrix of the flange coordinate system under attitude k2 relative to the base coordinate system.

7. The industrial robot TCP calibration method according to claim 3, characterized in that, It also includes the S5 quality assessment step, which evaluates the reliability of the attitude normal estimation, the risk of collinearity, the difference between two attitudes on the same plane, the overall calibration accuracy and / or cross-attitude consistency.

8. The industrial robot TCP calibration method according to claim 7, characterized in that, The reliability of attitude normal estimation is evaluated by the ratio of singular values ​​in SVD decomposition, i.e., the singular value of the normal / the singular value of the orthogonal direction in the plane. The risk of collinearity at sampling points is evaluated by the singular values ​​of the orthogonal directions in the plane / the singular values ​​of the main extension direction. The sufficiency of the difference in attitude angles between two objects on the same plane is evaluated by the condition number of the 3×3 linear system matrix. The overall calibration accuracy is evaluated by the root mean square of the residuals of the equations at each touch point. Cross-pose consistency is assessed by independently estimating the directed distance between two poses on the same plane.

9. The industrial robot TCP calibration method according to claim 6, characterized in that, The rotation matrix satisfies ||R_k1 -R_k2|| ≥ 0.005, and the angle between the Z-axis direction of the end tool and the plane normal is within the range of 0° to 60°.

10. The industrial robot TCP calibration method according to claim 3, characterized in that, In step S1, the number of data acquisition points for each pose shall be no less than 3 and shall not be collinear; the total number of data acquisition points shall be no less than 18. When the number of acquisition points for a certain attitude exceeds 3, the influence of measurement noise on normal estimation is reduced by using SVD least squares fitting; when the number of acquisition points for a certain attitude is equal to 3, the plane determined by the three points directly gives the plane normal.

Images