A Robot Joint Rotor Calibration Method Based on Freeform Surface Error Mapping
By using a method based on freeform surface error mapping, a joint screw error identification model is constructed by acquiring point clouds of the freeform surface of the robot end effector using a 3D scanner. This solves the problems of fitting error and absolute position dependence in visual calibration, and achieves high-precision and low-cost robot calibration.
Patent Information
- Application Number
- CN202610585520.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-29
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-04-29
AI Technical Summary
Existing vision-based robot kinematics calibration methods are susceptible to fitting errors and are highly dependent on high-precision absolute position, which limits calibration accuracy.
A method based on freeform surface error mapping is adopted. The point cloud of the freeform surface calibration object of the robot end is collected by a 3D scanner. By constructing a joint screw error identification model and iterative optimization, the residual of the point cloud normal distance is directly used as the optimization target, avoiding fitting error and getting rid of the dependence on high-precision absolute position.
It significantly improves calibration accuracy and stability, reduces calibration costs, enhances the accuracy of robot 3D reconstruction, and simplifies the operation process.
Smart Images

Figure CN122253206B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of robot calibration and computer vision, and in particular to a robot joint spinor calibration method based on freeform surface error mapping. Background Technology
[0002] As the core execution unit of intelligent manufacturing, the absolute positioning accuracy of industrial robots directly determines the quality of tasks such as assembly, processing, and measurement. However, due to factors such as manufacturing tolerances, long-term wear and tear, environmental temperature changes, and load deformation, the nominal kinematic model of a robot often deviates from its actual physical model, leading to a decrease in absolute positioning accuracy. Therefore, kinematic calibration of industrial robots, identifying and compensating for their geometric parameter errors, is a key technical means to ensure high-precision operation of robots.
[0003] Traditional robot calibration methods mainly rely on high-precision external measuring equipment, such as laser trackers. However, such equipment is expensive, complex to operate, and has harsh environmental requirements. It usually requires professional technicians to operate, resulting in high calibration costs and making it difficult to perform high-frequency, routine calibrations on the factory floor.
[0004] To reduce calibration costs, vision-based kinematic calibration methods have been developed in recent years. These methods utilize visual sensors such as cameras or 3D scanners to acquire the position or pose information of the robot's end effector, thereby identifying kinematic parameters. However, existing vision-based kinematic calibration methods generally suffer from the following shortcomings: (1) Accuracy loss due to spherical fitting: Existing methods typically install a standard sphere as a target at the robot's end effector, capture a point cloud of the sphere using a vision device, and obtain the coordinates of the sphere's center using a spherical fitting algorithm, which is then used as the measurement position of the robot's end effector. However, the point cloud data itself contains measurement noise, and the spherical fitting process further amplifies the noise effect, resulting in uncontrollable errors in the fitted sphere's center coordinates. This fitting error will ultimately be mapped to the identification results of kinematic parameters, making the calibration effect unstable and limiting accuracy.
[0005] (2) Dependence on high-precision absolute position: Most visual calibration methods are essentially improvements on laser tracker calibration methods, and their optimization goal is still to obtain the high-precision absolute position coordinates of the robot end effector in space. However, visual sensors are limited by their inherent field of view and imaging principles, making it difficult to simultaneously achieve high precision and a wide field of view in a large workspace, resulting in limited accuracy of the acquired absolute position information. Using such erroneous absolute poses as reference benchmarks for kinematic parameter identification fundamentally limits further improvement in calibration accuracy. Summary of the Invention
[0006] To address the issues of existing vision-based kinematic calibration methods being susceptible to fitting errors and having a strong dependence on high-precision absolute position, this application provides a robot joint screw calibration method based on freeform surface error mapping.
[0007] In a first aspect, this application provides a method for calibrating the screw quantity of a robot joint based on freeform surface error mapping, comprising the following steps: Install the freeform surface calibration object at the end of the robot; Control the robot to move to multiple poses, use a 3D scanner to collect point clouds of freeform surface calibration objects and record the robot's end-effector pose; Hand-eye calibration is performed based on the point cloud of the freeform surface calibration object to obtain the hand-eye matrix of the 3D scanner relative to the robot end effector; Using the robot's end-effector pose and hand-eye matrix, the point cloud of the freeform surface calibration object is transformed from the 3D scanner coordinate system to the robot's end-effector coordinate system to obtain the reconstructed point cloud; Construct nearest point pairs in the overlapping region of the reconstructed point cloud and calculate the normal distance residual for each nearest point pair; Construct a joint screw displacement error identification model; Based on the identification model of the normal distance residuals and joint screw error of each nearest point pair, a calibration objective function is established with the goal of minimizing the sum of squares of the normal distance residuals. Based on the calibration objective function, an error identification objective function is constructed and iteratively solved. The obtained joint screw error vector is compensated to the initial value of the joint screw, and the iteration continues until the change in the value of the calibration objective function is less than a preset threshold.
[0008] Furthermore, the steps for calculating the normal distance residuals for each nearest point pair include: Take two adjacent frames of point cloud with overlapping regions from the reconstructed point cloud according to the acquisition order, and use the nearest neighbor search algorithm to construct the nearest point pair between the two frames of point cloud. Calculate the normal vector at each point and unify the direction of the normal vectors to point to the calibration center of the freeform surface. Calculate the normal distance residual for each pair of closest points.
[0009] Furthermore, the construction method of the joint screw error identification model includes: establishing a robot exponential product positive kinematics model, performing differentiation and linearization on the robot exponential product positive kinematics model, and establishing a linear relationship between the robot matrix error components and the joint screw error vector, wherein the coefficient matrix in the linear relationship is defined as the identification Jacobian matrix.
[0010] Furthermore, the joint screw error vector is represented in the form of adjoint error, and the joint screw error vector is used as the input of the adjoint transformation to participate in the update of the joint screw, so as to avoid introducing orthogonalization calculation error during the iterative correction process.
[0011] Furthermore, the identification Jacobian matrix is composed of the adjoint transformation matrix and its coefficient matrix corresponding to each joint screw; wherein, the adjoint transformation matrix is used to transform the joint screw error from the joint coordinate system to the end-effector coordinate system, and the coefficient matrix is used to extract the screw Lie algebra coordinates.
[0012] Furthermore, the method for constructing the calibration objective function includes: establishing a linear relationship between the normal distance residuals of each nearest point pair and the joint screw error vector, and accumulating the sum of the squares of the normal distance residuals of all nearest point pairs as the calibration objective function to be minimized.
[0013] Further, the step of constructing an error identification objective function based on the calibration objective function and iteratively solving it, and compensating the obtained joint rotation error vector to the initial value of the joint rotation until the change in the value of the calibration objective function is less than a preset threshold includes: Based on the calibration objective function, a joint screw error vector identification objective function is constructed. The damped Newton method is used to solve the joint screw error vector identification objective function to obtain the joint screw error vector of the current iteration step. The joint screw error vector of the current iteration step is decomposed into multiple error components, and the joint screw of each joint is updated and compensated through the adjoint transformation. Repeat the iteration until the difference between the calibration objective function values of two consecutive iterations is less than a preset threshold, then stop the iteration.
[0014] Secondly, this application provides a robot joint rotation calibration device based on freeform surface error mapping, comprising: The data acquisition and calibration module is used to control the robot end effector to move to multiple poses, acquire the point cloud of the freeform surface calibration object installed on the robot end effector and record the corresponding robot end effector pose, and perform hand-eye calibration based on the point cloud of the freeform surface calibration object to obtain the hand-eye matrix of the 3D scanner relative to the robot end effector. The point cloud processing and residual calculation module is used to transform the point cloud of the freeform surface calibration object from the 3D scanner coordinate system to the robot end-eye coordinate system using the robot end pose and the hand-eye matrix to obtain the reconstructed point cloud, and to construct the nearest point pairs in the overlapping area of the reconstructed point cloud and calculate the normal distance residual of each nearest point pair. The model and objective function construction module is used to construct a joint screw error identification model and, based on the normal distance residual and the joint screw error identification model, establish a calibration objective function with the goal of minimizing the sum of squares of the normal distance residual; The iterative solution module is used to construct an error identification objective function based on the calibration objective function and iteratively solve it, compensate the obtained joint screw error vector to the initial value of the joint screw, and iterate until the change of the calibration objective function value is less than a preset threshold.
[0015] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned robot joint rotation calibration method based on freeform surface error mapping.
[0016] Fourthly, this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-mentioned method for robot joint rotation calibration based on freeform surface error mapping.
[0017] In summary, this application includes at least one of the following beneficial technical effects: This application uses a freeform surface as the calibration object, employs a 3D scanner to acquire multiple frames of point cloud data and record the robot's pose, and directly constructs an optimization target based on the point cloud normal distance residual. This avoids the accuracy loss caused by standard sphere fitting in traditional methods, improving accuracy by more than 10%, while also eliminating the reliance on high-precision absolute position references. Compared to laser tracker solutions, this application reduces calibration costs by more than 50% and simplifies the operation process.
[0018] This application establishes a joint screw quantity error correction model and iteratively optimizes it using the damped Newton method, enabling the joint screw quantity parameters to accurately converge to their theoretical values. Simultaneously, it introduces adjoint errors to replace traditional calibration micro-values, effectively avoiding the cumulative errors caused by orthogonalization calculations. Experimental data shows that after calibration, the robot's 3D reconstruction accuracy is improved by more than 50%, and the system stability is significantly enhanced. Attached Figure Description
[0019] Figure 1 This is a flowchart of a robot joint screw calibration method based on freeform surface error mapping according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the error minimization reconstruction principle in the embodiments of this application; Figure 3 This is a comparison of point cloud errors and joint errors before and after reconstruction in the embodiments of this application. Detailed Implementation
[0020] The following is in conjunction with the appendix Figure 1-3 This application will be described in further detail.
[0021] This application discloses a robot joint screw calibration method based on freeform surface error mapping. Its hardware system includes: an industrial robot, a 3D scanner fixedly installed in the robot's workspace, and a freeform surface calibration object installed at the robot's end effector. The freeform surface calibration object is an arbitrary irregular surface object with continuous curvature, its centroid located inside the freeform surface calibration object, and the 3D model of the freeform surface calibration object is known, with its base coordinate system coinciding with the robot's end effector flange coordinate system.
[0022] Reference Figure 1 The robot joint screw calibration method based on freeform surface error mapping includes the following steps: Step 1: Install the freeform surface calibration object onto the robot's end effector; Step 2: Control the robot to move to multiple poses, use a 3D scanner to collect point clouds of freeform surface calibration objects and record the robot's end-effector pose; Step 3: Perform hand-eye calibration based on the point cloud of the freeform surface calibration object to obtain the hand-eye matrix of the 3D scanner relative to the robot end effector; Step 4: Using the robot's end-effector pose and hand-eye matrix, transform the point cloud of the freeform surface calibration object from the 3D scanner coordinate system to the robot's end-effector coordinate system to obtain the reconstructed point cloud; Step 5: Construct nearest point pairs in the overlapping region of the reconstructed point cloud and calculate the normal distance residual for each nearest point pair; Step 6: Construct a joint rotation error identification model; Step 7: Based on the identification model of the normal distance residuals and joint screw error of each nearest point pair, establish a calibration objective function with the goal of minimizing the sum of squares of the normal distance residuals; Step 8: Construct an error identification objective function based on the calibration objective function and solve iteratively. Compensate the obtained joint screw error vector to the initial value of the joint screw, and iterate until the change in the calibration objective function value is less than a preset threshold.
[0023] This application proposes a robot joint screw quantity calibration method based on freeform surface error mapping, aiming to achieve kinematic parameter identification of industrial robots with low cost and simple operation. This method only requires a 3D scanner to acquire multiple point clouds of an arbitrary freeform surface object at multiple positions and postures at the robot's end effector, and can automatically complete the calibration, making it particularly suitable for rapid calibration of robots in long-term service in factory environments. Its core lies in constructing a joint screw quantity error correction model using freeform surface reconstruction error mapping, directly using the point cloud normal distance residual as the optimization objective. This avoids the accuracy loss introduced by fitting a standard sphere center in traditional visual calibration and eliminates the dependence on high-precision absolute position measurement. By iteratively minimizing the reconstruction error using the damped Newton method, and introducing adjoint errors to replace traditional calibration micro-quantities, not only is the computational error caused by orthogonalization avoided, but the stability and accuracy of calibration are also significantly improved, ultimately obtaining stable and high-precision calibration results.
[0024] In step 1, the freeform surface calibration object is installed on the end effector of the robot, and the 3D scanner is fixed so that the freeform surface calibration object is within the field of view of the 3D scanner.
[0025] In step 2, the robot is controlled to move to multiple poses, and a 3D scanner is used to collect point clouds of freeform surface calibration objects and record the robot's end-effector pose.
[0026] When controlling the robot to move to multiple poses, all joints of the robot should be displaced, and the position of the robot's end effector should fill the entire field of view of the 3D scanner. It is recommended to use linear motion to move the end effector position, and then use repositioning motion to rotate each axis. At least 30 frames of point cloud data should be collected to prevent matrix degradation.
[0027] In step 3, hand-eye calibration is performed based on the point cloud of the freeform surface calibration object to obtain the hand-eye matrix of the 3D scanner relative to the robot end effector.
[0028] The hand-eye calibration adopts a globally optimal hand-eye calibration method based on point cloud registration (for example, a globally optimal robust hand-eye calibration method disclosed in patent publication number CN118046386A) to obtain the hand-eye matrix that minimizes the registration error of all point clouds.
[0029] In step 4, the point cloud is transformed from the 3D scanner coordinate system to the robot end-effector coordinate system using the robot end-effector pose and hand-eye matrix to obtain the reconstructed point cloud.
[0030] The specific calculation formula for constructing the reconstructed point cloud is as follows: Formula (1) In formula (1), Let represent the homogeneous coordinates of the i-th frame point cloud in the scanner coordinate system. This represents the homogeneous coordinates of the point cloud in the i-th frame in the robot's end effector coordinate system. This represents the hand-eye matrix obtained in step 3. This represents the robot end-effector pose matrix corresponding to the i-th image capture, and all of them... Convert to This allows for the reconstruction of the point cloud of the freeform surface calibration object, such as... Figure 2 As shown.
[0031] In step 5, nearest point pairs are constructed in the overlapping regions of the reconstructed point cloud, and the normal distance residuals of each nearest point pair are calculated. Specific steps include: Step 5.1: Take the i-th frame of the reconstructed point cloud with overlapping regions in the acquisition order. and the point cloud of frame i+1 Using KD-tree structure for search Each point in exist The nearest point in the middle Form the nearest point pair .
[0032] Step 5.2: Calculate using principal component analysis. Each point in Corresponding normal vector The subscript k represents the index of the nearest point in the dense model point cloud, and the normal vector direction uniformly points to the calibration center of the freeform surface.
[0033] Step 5.3: Calculate the normal distance residual for each pair of closest points. The residual is the projected distance of the line connecting the points onto the direction of the normal vector: Formula (2) In formula (2), for transpose, superscript i For point cloud frame count, subscript j The index is the number of a point in the point cloud.
[0034] In step 6, a joint rotation error identification model is constructed. The specific steps include: The robot's exponential product positive kinematics model is differentiated and linearized to establish a linear relationship between the robot matrix error components and the joint screw error vector. The coefficient matrix in the linear relationship is defined as the identification Jacobian matrix. Specifically: The exponential product positive kinematics model of the six-DOF robot is established as follows: Formula (3) In formula (3), M This is the end effector matrix when the robot is in its initial pose.e [Si]θi For the first i Rigid body transformation matrices corresponding to each joint Represents the set of joint angles for each axis. These are the joint angles of joints 1 through 6 of a six-degree-of-freedom robot. For the first i Joint rotation Lie algebra form: Formula (4) In formula (4), Let i be the linear velocity of the i-th joint. For the first i Joint angular velocity The Lie algebraic form of it is a 3×3 antisymmetric matrix. Specifically, for moving joints, Then the corresponding Indicates the distance traveled.
[0035] The methods for constructing a joint rotation error identification model include: Formula (3) for Differentiating and right-multiplying by its inverse matrix yields the robot matrix error components: Formula (5) In formula (5), , Let be the set of 6th-order square matrices (6 rows and 6 columns) consisting of real number elements. This is the robot's end-effector pose matrix. for The differential vector, This is the joint structure error component of the robot. After linearization, this component becomes the joint screw error vector.
[0036] Furthermore, linearize equation (5): Formula (6) In formula (6), This represents the direct mapping transformation between Lie algebra elements and their corresponding spinors. This represents the screw quantity error vector matrix for each joint. These are the screw error vectors for the 1st to 6th joints of a six-DOF robot; This represents the joint screw error vector to be solved. These are the screw error vectors to be solved for the 1st to 6th joints of a six-DOF robot. It represents the set of 36-dimensional column vectors (36 rows and 1 column matrices) consisting of real number elements.
[0037] The joint screw error vector is represented in the form of adjoint error. That is, the error vector is used as the input of the adjoint transformation to participate in the update of the joint screw, so as to avoid introducing orthogonalization calculation error in the iterative correction process and ensure that the updated joint screw always satisfies the Lie algebra constraint.
[0038] In formula (6), J The identification Jacobian matrix is composed of the adjoint transformation matrix and its coefficient matrix corresponding to each joint screw quantity. The adjoint transformation matrix is used to transform the joint screw quantity error from the joint coordinate system to the end-effector coordinate system, and the coefficient matrix is used to extract the screw quantity Lie algebra coordinates. Formula (7) In formula (7), Represents the adjoint computation of the matrix. For joint rotation, For joint angles, where the subscript is... k The factorial index. k Not greater than i -1, i Joint number; for The coefficient matrix is calculated as follows: Formula (8) In formula (8), i i For the first i The joint angles of each joint. I 6 represents a 6-dimensional identity matrix; P i For the first i The total rotation angle of each joint around its joint axis, Ω i For the first i The antisymmetric matrix corresponding to the 6-dimensional Lie algebra of each joint is calculated as follows: Formula (9) In formula (9), oh 1i , oh 2i , oh 3i They are vectors oh i The three components.
[0039] In formula (7), for The coefficient matrix is calculated as follows: Formula (10) In formula (10), i ηiΩ represents the total rotation angle (magnitude of the rotation vector) of the end transformation. ηi The 6-dimensional Lie algebra antisymmetric matrix of the terminal transformation is calculated as follows: Formula (11) In formula (11), oh 1ηi , oh 2ηi , oh 3ηi They are vectors oh ηi The three components, oh ηi For vectors or i rotation vector, v ηi For vectors or i The translation vector.
[0040] In formula (7), The coefficient matrix is 6×6, and its calculation method is as follows: Formula (12) In formula (12), oh Ad (e [ηi] )Si For vectors rotation vector, v Ad (e [ηi] )Si For vectors Translation vector, for and The product of A 6×6 vector is The adjoint vector, for or i The corresponding rigid body transformation matrix.
[0041] In step 7, based on the identification model of the normal distance residuals and joint screw error of each nearest point pair, a calibration objective function is established with the goal of minimizing the sum of squared normal distance residuals. Specific steps include: Assumption These are the robot's rotation error vector and translation error vector, respectively. If we represent three-dimensional real space, then: Formula (13) Assumption Let the coordinates of the same point be in the robot's base coordinate system and the robot's end effector coordinate system, respectively. Substituting the robot error and ignoring the second-order differential term, we have: Formula (14) In formula (14), yes The error is minute. It is a robot rotation matrix. It is the robot's translation vector. I It is a 3×3 identity matrix, which, when expanded, yields: Formula (15) In formula (15), x Let be the joint rotation error vector to be solved. K Given a coefficient matrix of 3×36, substituting formula (15) into formula (2) yields: Formula (16) In formula (16), For point Time normal distance, The coefficient matrix, It is a 3×36 coefficient matrix, that is, from the first... i In the point cloud of the first frame j Calculated from points K matrix; It is a 3×36 coefficient matrix, that is, from the first... i +1 frame point cloud j Calculated from points K matrix.
[0042] The calibration objective function is established by first squaring and then summing the residuals of the normal distances to the nearest points: Formula (17) In formula (17), n The total number of frames in the point cloud. m i This represents the total number of points in a single frame of the point cloud.
[0043] In step 8, an error identification objective function is constructed based on the calibration objective function and solved iteratively. The obtained joint rotation error vector is compensated to the initial value of the joint rotation. The iteration continues until the change in the calibration objective function value is less than a preset threshold. The specific steps include: Step 8.1, Error Calculation: Construct and solve a system of linear equations containing damping coefficients based on the calibration objective function to obtain the joint screw error vector for the current iteration step. Specifically, based on formulas (16) and (17), expand to obtain the joint screw error vector identification objective function: Formula (18) The joint screw error vector is obtained by solving formula (18) using the damped Newton method: Formula (19) In formula (19), I It is the identity matrix. The damping coefficient is taken as in this method. The damped Newton method can make the convergence of the target function for identifying the joint screw error vector more stable.
[0044] Step 8.2, Error Compensation: Decompose the joint screw error vector of the current iteration step into error components corresponding to each joint, and update and compensate the joint screw of each joint through the adjoint transformation. Specifically: Will x Decomposed into There are 6 error components, corresponding to the screw rotation errors of joints 1 to 6 respectively. Then, the joint screw rotation is compensated and updated using the following formula. : Formula (20) In formula (20), the joint screw is updated using the accompanying error. On the one hand, it corresponds to formula (6); on the other hand, the accompanying error ensures that the joint screw always satisfies the orthogonal constraint during the update process.
[0045] Repeat the iteration until the difference between the calibration objective function values of two consecutive iterations is less than a preset threshold of 10. -5 Stop iterating.
[0046] This embodiment uses a 3D vision scanner as the calibration device, reducing calibration costs by more than 50% compared to a laser tracker. This application uses point cloud normal distance error as the optimization target, eliminating the fitting process of traditional methods and further improving calibration accuracy by more than 10%. Figure 3 As shown, the robot reconstruction accuracy can be improved by more than 50% before and after calibration, which has high practical value.
[0047] This embodiment provides a robot joint screw calibration device based on freeform surface error mapping, including: The data acquisition and calibration module is used to control the robot end effector to move to multiple poses, acquire the point cloud of the freeform surface calibration object installed on the robot end effector and record the corresponding robot end effector pose, and perform hand-eye calibration based on the point cloud of the freeform surface calibration object to obtain the hand-eye matrix of the 3D scanner relative to the robot end effector. The point cloud processing and residual calculation module is used to transform the point cloud of the freeform surface calibration object from the 3D scanner coordinate system to the robot end-effector coordinate system using the robot end pose and hand-eye matrix to obtain the reconstructed point cloud, and to construct the nearest point pairs in the overlapping area of the reconstructed point cloud and calculate the normal distance residual of each nearest point pair. The model and objective function construction module is used to construct the joint screw error identification model and, based on the normal distance residual and the joint screw error identification model, establish a calibration objective function with the goal of minimizing the sum of squares of the normal distance residual; The iterative solution module is used to construct an error identification objective function based on the calibration objective function and solve iteratively. The obtained joint screw error vector is compensated to the initial value of the joint screw, and the iteration continues until the change in the calibration objective function value is less than a preset threshold.
[0048] This embodiment also provides an electronic device, which may include a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other through the communication bus. The processor can call logical instructions in the memory to execute the above-described robot joint rotation calibration method based on freeform surface error mapping.
[0049] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0050] This embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the above-described robot joint spinor calibration method based on freeform surface error mapping.
[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for calibrating the screw quantity of a robot joint based on freeform surface error mapping, characterized in that: Includes the following steps: Install the freeform surface calibration object at the end of the robot; Control the robot to move to multiple poses, use a 3D scanner to collect point clouds of freeform surface calibration objects and record the robot's end-effector pose; Hand-eye calibration is performed based on the point cloud of the freeform surface calibration object to obtain the hand-eye matrix of the 3D scanner relative to the robot end effector; Using the robot's end-effector pose and hand-eye matrix, the point cloud of the freeform surface calibration object is transformed from the 3D scanner coordinate system to the robot's end-effector coordinate system to obtain the reconstructed point cloud; In the overlapping region of the reconstructed point cloud, the nearest point pair is constructed, and the normal distance residual of each nearest point pair is calculated. This includes: taking two adjacent frames of point cloud with overlapping regions in the reconstructed point cloud according to the acquisition order, constructing the nearest point pair between the two frames of point cloud using the nearest neighbor search algorithm; calculating the normal vector of each point and unifying the direction of the normal vector to point to the calibration center of the freeform surface; and calculating the normal distance residual of each pair of nearest points. Construct a joint screw displacement error identification model; Based on the identification model of the normal distance residuals and joint screw error of each nearest point pair, a calibration objective function is established with the goal of minimizing the sum of squares of the normal distance residuals. The process includes: constructing an error identification objective function based on the calibration objective function and iteratively solving it; compensating the obtained joint screw error vector to the initial value of the joint screw; iterating until the change in the value of the calibration objective function is less than a preset threshold; including: constructing a joint screw error vector identification objective function based on the calibration objective function; solving the joint screw error vector identification objective function using the damped Newton method to obtain the joint screw error vector of the current iteration step; decomposing the joint screw error vector of the current iteration step into multiple error components; updating and compensating the joint screw of each joint through the adjoint transformation; repeating the iteration until the difference between the calibration objective function values of two consecutive iterations is less than a preset threshold, at which point the iteration stops.
2. The robot joint screw calibration method based on free-form surface error mapping according to claim 1, characterized in that: The construction method of the joint screw error identification model includes: establishing a robot exponential product positive kinematics model, performing differentiation and linearization on the robot exponential product positive kinematics model, and establishing a linear relationship between the robot matrix error components and the joint screw error vector. The coefficient matrix in the linear relationship is defined as the identification Jacobian matrix.
3. The robot joint screw calibration method based on free-form error mapping of claim 2, wherein: The joint screw error vector is represented in the form of adjoint error. The joint screw error vector is used as the input of the adjoint transformation to participate in the update of the joint screw, so as to avoid introducing orthogonalization calculation error during the iterative correction process.
4. The robot joint screw calibration method based on free-form surface error mapping of claim 2, wherein: The identification Jacobian matrix is composed of the adjoint transformation matrix and its coefficient matrix corresponding to each joint screw; wherein, the adjoint transformation matrix is used to transform the joint screw error from the joint coordinate system to the end-effector coordinate system, and the coefficient matrix is used to extract the screw Lie algebra coordinates.
5. The robot joint screw calibration method based on free-form error mapping of claim 1, wherein: The calibration objective function is constructed by establishing a linear relationship between the normal distance residuals of each nearest point pair and the joint screw error vector, and summing the squares of the normal distance residuals of all nearest point pairs as the calibration objective function to be minimized.
6. A robot joint wrench calibration apparatus based on free-form surface error mapping, characterized by: include: The data acquisition and calibration module is used to control the robot end effector to move to multiple poses, acquire the point cloud of the freeform surface calibration object installed on the robot end effector and record the corresponding robot end effector pose, and perform hand-eye calibration based on the point cloud of the freeform surface calibration object to obtain the hand-eye matrix of the 3D scanner relative to the robot end effector. The point cloud processing and residual calculation module is used to transform the point cloud of the freeform surface calibration object from the 3D scanner coordinate system to the robot end-effector coordinate system using the robot end-effector pose and the hand-eye matrix to obtain a reconstructed point cloud. It then constructs nearest point pairs in the overlapping regions of the reconstructed point clouds and calculates the normal distance residual for each nearest point pair. This includes: taking two adjacent frames of point clouds with overlapping regions in the reconstructed point cloud according to the acquisition sequence; constructing nearest point pairs between the two frames using a nearest neighbor search algorithm; calculating the normal vector of each point and unifying the direction of the normal vectors to point towards the center of the freeform surface calibration object; and calculating the normal distance residual for each pair of nearest point pairs. The model and objective function construction module is used to construct a joint screw error identification model and, based on the normal distance residual and the joint screw error identification model, establish a calibration objective function with the goal of minimizing the sum of squares of the normal distance residual; The iterative solution module is used to construct an error identification objective function based on the calibration objective function and iteratively solve it, compensating the obtained joint screw error vector to the initial value of the joint screw, and iterating until the change in the value of the calibration objective function is less than a preset threshold. This includes: constructing a joint screw error vector identification objective function based on the calibration objective function; solving the joint screw error vector identification objective function using the damped Newton method to obtain the joint screw error vector of the current iteration step; decomposing the joint screw error vector of the current iteration step into multiple error components, and updating and compensating the joint screw of each joint through an adjoint transformation; repeating the iteration until the difference between the calibration objective function values of two consecutive iterations is less than a preset threshold, at which point the iteration stops.
7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that: When the processor executes the program, it implements a robot joint spinor calibration method based on freeform surface error mapping as described in any one of claims 1-5.
8. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that: When the computer program is executed by the processor, it implements a robot joint spinor calibration method based on freeform surface error mapping as described in any one of claims 1-5.
Citation Information
Patent Citations
Global optimal robust hand-eye calibration method and device
CN118046386A
Hand-eye calibration method and device for picking robot based on TOF camera
CN116021519A
Robot-assisted multi-view three-dimensional scanning measurement method
CN116309879A
