Robot calibration automatic alignment method based on visual and PSD combined guidance

Abstract

This invention relates to an automated alignment method for robot calibration based on vision and PSD joint guidance, encompassing fields such as automated calibration of industrial robots and laser precision alignment technology. It addresses the shortcomings of existing four-PSD calibration methods, which rely on manual labor and are inefficient, as well as the limitations of existing vision-PSD hybrid servo technologies, which can only achieve single-point alignment, cannot support time-sharing projection of multiple PSDs from the same laser, and cannot construct dual-optical-path non-planar geometric constraints. The invention achieves this through steps including: pre-calibration of the calibration system and multi-coordinate system; two projections, using the first and second optical path measurement groups respectively to achieve joint alignment of vision and PSD; precise alignment with dual-optical-path non-planar spatial geometric constraints; and fully automated multi-station data acquisition and calibration parameter solving. This invention, through a vision-PSD-vision-PSD multi-round collaborative and time-sharing dual-optical-path non-planar constraint architecture, is adaptable to various industrial robot laser precision calibration scenarios.

Description

Technical Field

[0001] This invention relates to the fields of automated calibration and laser precision alignment of industrial robots, specifically to an automatic alignment method for robot calibration based on vision and PSD joint guidance. This method is applicable to scenarios such as laser processing, laser inspection, and high-precision calibration of end tools for industrial robots, and is especially suitable for high-precision calibration of robot geometric parameters under a four-PSD co-constraint architecture. Background Technology

[0002] Currently, using a fixed laser at the end effector of an industrial robot in conjunction with a position sensitive detector (PSD) for robot geometric parameter calibration is one of the core methods in the field of high-precision robot error identification. This has significant theoretical and engineering implications for improving the robot kinematic error compensation system and expanding the high-precision application scenarios of industrial robots. Current high-precision robot laser calibration generally adopts a four-PSD co-constraint calibration architecture. Utilizing the geometric constraint relationship between two sets of non-planar laser optical paths, it can accurately identify robot kinematic parameters and laser beam installation deviation parameters, achieving significantly higher calibration accuracy than single-sensor calibration methods.

[0003] Existing multi-PSD calibration technology has obvious technical shortcomings: traditional calibration relies entirely on manual visual inspection of light spots and manual fine-tuning of robot posture to complete centering one by one. At the same time, manual verification of optical path geometric constraints is required. The operation is cumbersome, time-consuming, has large human error, and poor calibration consistency, which cannot meet the needs of automated and standardized batch calibration.

[0004] In recent years, a hybrid control strategy combining visual servoing and PSD feedback has been introduced into the field of robotic laser alignment. Existing technologies have proposed a two-stage hybrid control method based on visual coarse servoing and PSD fine servoing, achieving automatic alignment of a single laser spot with a single PSD target under single-projection conditions. However, this type of existing technology has the following inherent drawbacks:

[0005] 1. Only supports single-point alignment: The existing vision-PSD hybrid servo method only solves the problem of aligning one laser spot with one PSD target. The controlled object is the two-dimensional planar position deviation of a single spot, which cannot meet the requirement of aligning two sets of optical paths separately in four PSD co-constraint calibration.

[0006] 2. No time-sharing multi-target projection involved: In existing technologies, visual guidance and PSD closed loop are only executed once. There has never been a publicly disclosed or implied technical concept of projecting the same laser onto multiple PSD targets in a time-sharing manner and constructing spatial geometric constraints based on multiple projection data.

[0007] 3. Lack of dual-path spatial geometric constraint mechanism: The existing scheme only focuses on the two-dimensional position deviation of the light spot and does not introduce the spatial distance between the two optical paths as a constraint condition. This makes it easy to produce pseudo-qualified samples where the single-path light spot is centered and the overall optical path spatial configuration does not meet the modeling conditions, resulting in ill-conditioned calibration equations and model failure.

[0008] 4. The coordination method between vision and PSD is singular: In the existing technology, the switching between vision and PSD is only a single unidirectional switch from vision window to PSD lock, without involving a multi-round alternating projection and time-sharing dominant coordination mechanism of vision-PSD-vision-PSD.

[0009] More importantly, current technology lacks a reasonable technical path and motivation for time-division projection using a single laser: In existing technologies, if multiple lasers are used to project onto multiple PSDs separately, the differences in the emission origin and beam directionality between different lasers will introduce additional system errors, and the collaborative control of multiple lasers is highly complex and costly. If a single laser is used, the existing vision-to-PSD single-switching architecture cannot solve the problems of cross-group navigation, time-division alignment, and spatial constraint unification for the same light source across multiple workstations. In other words, existing technologies neither disclose the specific implementation path of time-division projection of a single laser onto multiple PSDs, nor do they realize that a dual-optical-path non-planar constraint must be introduced after time-division projection to avoid the deep-seated technical problem of ill-conditioned calibration equations.

[0010] In summary, there is an urgent need for a fully automated, efficient, and high-precision robotic laser beam alignment method that can project the same laser onto multiple PSDs in a time-sharing manner and be guided in stages by vision and PSDs during each projection. Summary of the Invention

[0011] This invention addresses the shortcomings of existing four-PSD calibration methods, such as reliance on manual labor and low efficiency, as well as the limitations of existing vision-PSD hybrid servo technologies, which can only achieve single-point alignment, cannot support time-sharing projection of multiple PSDs from the same laser, and cannot construct dual-optical-path non-planar geometric constraints. It provides an automated alignment method for robot calibration based on joint vision and PSD guidance. This method executes vision-guided long-distance windowing and precise PSD closed-loop locking relatively independently in two time-sharing projections. Furthermore, it introduces dual-optical-path non-planar spatial geometric constraint coupling optimization to achieve time-sharing alignment of four PSDs from the same laser source, ensuring the spatial geometric validity of each calibration sample and fundamentally solving the problem of calibration model failure.

[0012] An automatic alignment method for robot calibration based on vision and PSD joint guidance is implemented by the following steps:

[0013] Step 1: Construct a calibration system to achieve pre-calibration of multiple coordinate systems and build a pre-calibrated four-coordinate system mapping model;

[0014] Step 2: First projection, using the first optical path measurement group to achieve joint alignment of vision and PSD;

[0015] The robot is controlled to move to the first preset station. The laser spot is guided into the window from a distance by vision. Then, the first set of double PSD precision alignment is completed through PSD closed-loop control, and the first set of data is stored.

[0016] Step 3: Second projection, using the second optical path measurement group to achieve joint alignment of vision and PSD, and storing the second set of data;

[0017] The laser continues to work, controlling the robot to switch to the second preset station. After vision cross-group navigation, the second set of dual PSD precision alignment is completed again through PSD closed-loop control, and the second set of data is stored.

[0018] Step 4: Based on the first set of data obtained in Step 2 and the second set of data obtained in Step 3, perform precise alignment using dual-optical-path non-plane spatial geometric constraints.

[0019] Based on two projections, a non-plane laser optical path is established, the straight-line distance between the two optical paths is calculated, a multi-target coupled evaluation function is constructed, and the dual optical paths are optimized using a gradient descent iterative algorithm. A dual composite constraint threshold is set: a spot plane alignment threshold and a non-plane distance deviation threshold. The dual composite constraint is automatically verified after each iteration. When the center deviation of the two spot paths and the non-plane straight-line distance deviation both meet the standards, the fine alignment is determined to be completed.

[0020] Step 5: Automatically collect multiple sets of posture data to complete the robot calibration and automatic alignment.

[0021] The beneficial effects of this invention are:

[0022] 1. The automatic alignment method for robot calibration described in this invention adopts a single-laser time-division projection and dual-optical-path spatial virtual modeling calibration alignment architecture: Unlike existing technologies that can only achieve single-point alignment, this invention projects the same laser onto two sets of dual PSDs in a time-division manner, and reconstructs spatial non-planar linear constraints based on the measured data of the two projections, fundamentally ensuring the geometric validity of each set of calibration samples, and filling the application gap of existing vision and PSD hybrid servo technology in the field of four PSD co-constraint calibration.

[0023] 2. The automatic alignment method for robot calibration described in this invention adopts a multi-round joint guidance mechanism of vision-PSD-vision-PSD: In the prior art, vision and PSD only switch once in one direction. In this invention, vision is used to enter the window and lock the PSD in the first projection. Vision is used to enter the window and lock the PSD again in the workstation switching. Vision is used to enter the window and lock the PSD again in the second projection. This forms a deep joint guidance mode in which vision and PSD are alternately activated and time-division dominated in multiple rounds, ensuring that the same laser beam can independently and accurately complete the window entry and centering in both projections.

[0024] 3. The robot calibration automatic alignment method described in this invention has a high degree of automation: no manual posture adjustment, visual alignment and manual verification are required throughout the entire process. It realizes unmanned operation of the entire process of spot search, window entry, coarse convergence, precise locking, cross-group navigation, spatial constraint optimization, data acquisition and workstation switching, eliminating human error and significantly improving calibration consistency.

[0025] 4. The automatic alignment method for robot calibration described in this invention balances speed and accuracy: relying on visual guidance to quickly complete spot addressing and cross-group switching, the calibration efficiency is improved by more than 3 times, and a single calibration takes ≤15 minutes; combined with PSD high-frequency precision feedback and dual-optical-path geometric hard constraints, the spot alignment accuracy can reach ±0.1mm, and the non-plane distance constraint accuracy is ≤0.2mm.

[0026] 5. The automatic alignment method for robot calibration described in this invention fundamentally ensures the good state of the calibration equations and eliminates model failure: Traditional single-alignment schemes, lacking dual-optical-path spatial geometric constraints, result in highly degraded spatial configurations of the collected samples, even approaching coplanar / parallel configurations. This leads to an extremely large condition number and severely ill-conditioned matrices in the kinematic calibration equations, causing divergence or distortion in the weighted least squares identification results. This invention uses the distance between skew lines as a hard constraint threshold parallel to the light spot coordinates, ensuring that each set of collected samples satisfies the spatial non-coplanar geometric configuration condition. This significantly reduces the condition number of the equations and guarantees the numerical stability and uniqueness of the solution in the weighted least squares identification algorithm. In other words, this invention solves not only the optimization problem of accuracy but also the fundamental problem of whether the calibration model is feasible, addressing the long-standing industry pain point of model failure in the four-PSD co-constraint calibration field.

[0027] 6. The automatic alignment method for robot calibration described in this invention has strong versatility: it does not require an open control interface for the robot, is compatible with various enclosed industrial robots, and is suitable for various laser precision calibration scenarios for industrial robots. The calibration efficiency of this invention is improved by more than 3 times, the sample constraint pass rate is 100%, and it is suitable for various laser precision calibration scenarios for industrial robots. Attached Figure Description

[0028] Figure 1This is a flowchart of the robot calibration automatic alignment method based on vision and PSD joint guidance as described in this invention;

[0029] Figure 2 This is a schematic diagram of the hardware system in the robot calibration automatic alignment method based on vision and PSD joint guidance described in this invention.

[0030] Figure 3 This is a schematic diagram of the mapping relationship between multiple coordinate systems;

[0031] Figure 4 A sequence diagram of multi-round joint guidance control for vision-PSD-vision-PSD;

[0032] Figure 5 A schematic diagram illustrating the principle of time-division projection of a dual-optical-path using a single laser.

[0033] Figure 6 This is a block diagram of the four-PSD dual-channel synchronous constraint precision alignment control principle.

[0034] In the diagram: 1. Industrial robot; 2. Laser; 3. Monocular industrial camera; 4. Four-PSD calibration device; 5. First PSD; 6. Second PSD; 7. Third PSD; 8. Fourth PSD; 9. First reflector; 10. Second reflector; 11. Data acquisition unit. Detailed Implementation

[0035] Specific Implementation Method 1: Combination Figures 1 to 6 This embodiment describes an automatic alignment method for robot calibration based on vision and PSD joint guidance. This embodiment uses a general-purpose six-axis industrial robot as the calibration body and is applied to industrial conditions in a typical workshop with normal temperature and low light interference. It achieves the following process: calibration system setup and multi-coordinate system pre-calibration, first projection; vision and PSD joint alignment using a first optical path measurement group; second projection; and vision and PSD joint alignment using a second optical path measurement group. The process also includes dual-optical-path non-plane spatial geometric constraint precision alignment, fully automatic multi-station data acquisition, and calibration parameter solution. Figure 1 As shown.

[0036] The automatic alignment method for robot calibration described in this embodiment is implemented by the following steps:

[0037] Step 1. Calibration system setup and pre-calibration of the multi-coordinate system. The specific implementation process is as follows:

[0038] Step 11. Calibration system setup;

[0039] like Figure 2 As shown, a calibration system is constructed consisting of an industrial robot 1, a laser 2, a monocular industrial camera 3, a four-PSD calibration device 4, and a data acquisition unit 11.

[0040] Laser 2 is rigidly locked to the sixth axis flange of the end of industrial robot 1 using a flange clamp to ensure that there is no relative shaking between laser 2 and the end of the robot, and to avoid beam deviation during movement; monocular industrial camera 3 is locked and fixed directly above the center of the four PSD calibration device 4 using a bracket, and to ensure that the optical axis of monocular industrial camera 3 is perpendicular to the reference plane of the calibration device 4, with no relative displacement or vibration deviation.

[0041] In this embodiment, the industrial robot 1 is a general-purpose six-axis industrial robot with a repeatability of ±0.05mm; the end effector of the industrial robot 1 is equipped with a 650nm red semiconductor laser with a spot diameter of 0.5mm and continuously adjustable power.

[0042] The monocular industrial camera 3 is a fixed monocular industrial camera with a resolution of 2 million pixels and a frame rate of 30fps. It is equipped with a 16mm fixed-focus industrial lens, and its field of view covers the entire working surface of the four PSD calibration device 4.

[0043] The four-PSD calibration device 4 has four built-in high-precision PSD targets, namely: first PSD5, second PSD6, third PSD7 and fourth PSD8; and two high-definition first reflective mirrors 9 and second reflective mirrors 10 placed above the working surface of the four-PSD calibration device 4, with each reflective mirror facing downwards and the mirror surfaces parallel to the working surface of the four-PSD calibration device 4; each high-precision PSD target has an effective photosensitive area of ​​5mm×5mm, a measurement resolution of 0.01mm, a response rate of 1kHz, and is equipped with a physical light-shielding structure and a narrow-band filter structure to filter out ambient stray light interference;

[0044] The data acquisition unit 11 integrates a high-speed data acquisition card to achieve synchronous acquisition of visual images, PSD analog signals, and robot joint data.

[0045] In this embodiment, the first PSD5, the first reflector 9 and the second PSD6 constitute the first optical path measurement group, and the third PSD7, the second reflector 10 and the fourth PSD8 constitute the second optical path measurement group.

[0046] Step 12. Multi-coordinate system pre-calibration, such as... Figure 3 As shown;

[0047] Step 121. Use a 9×12 high-precision ceramic checkerboard calibration plate to complete the intrinsic parameter calibration of the monocular industrial camera 3. Collect 20 sets of checkerboard images at different angles and distances. Solve for the camera focal length, principal point coordinates, radial distortion and tangential distortion coefficients using Zhang's calibration method to complete image distortion correction and eliminate lens imaging errors.

[0048] Step 122. Based on the multi-point pose matching of the calibration board, fit the fixed extrinsic homogeneous transformation matrix of the camera coordinate system relative to the global coordinate system of the four PSD calibration device, and establish the mapping relationship between the camera coordinate system and the global coordinate system of the four PSD calibration device.

[0049] Step 123. Control the industrial robot 1 to complete 16 sets of spatially discrete and posture-differentiated teaching poses in sequence. Under each pose, collect the precise coordinates of the laser spot on the first PSD5, the second PSD6, the third PSD7 and the fourth PSD8. Combine the installation structure relationship between the robot 1 and the laser 2, and obtain the mapping relationship between the robot coordinate system and the PSD measurement coordinate system through the least squares algorithm iterative fitting.

[0050] Step 124. Based on the inherent mapping relationship between the global coordinate system of the four PSD calibration device and the PSD measurement coordinate system determined by the spatial structure, establish a pre-calibrated four-coordinate system mapping model consisting of the robot coordinate system, the global coordinate system of the four PSD calibration device, the PSD measurement coordinate system and the camera coordinate system, and realize bidirectional accurate calculation between any two coordinate systems within the four coordinate systems.

[0051] Step 2. First projection: Use the first optical path measurement group to achieve joint alignment of vision and PSD, and store the first set of data;

[0052] like Figure 4 As shown, the industrial robot 1 is controlled to move from an initial arbitrary pose to the first preset workstation. The laser 2 is activated with low-power regulated output to ensure uniform laser spot brightness, without overexposure or dark spots. The industrial robot 1 is driven to complete the entire process of projecting the laser beam onto the first PSD5, reflecting it from the first PSD5 to the first reflector 9, and then reflecting it again from the first reflector 9 to the second PSD6. During this projection process, vision-guided entry and PSD closed-loop precision alignment work together, as detailed below:

[0053] Step 21. Visually guide the first projection from a distance to achieve laser spot entry into the window;

[0054] A monocular industrial camera 3 continuously acquires raw images projected onto the workspace at a frame rate of 30fps. The raw images are preprocessed, sequentially performing grayscale conversion, Gaussian noise reduction, threshold binarization, and morphological opening denoising to accurately extract the high-contrast laser spot contour and its position information in the workspace. The sub-pixel center coordinates of the laser spot are calculated using contour moments. Based on the pre-calibrated four-coordinate system mapping model obtained in step 124, the distance between the current laser spot and the effective photosensitive areas of the first PSD5 and the second PSD6 is calculated. Segmented smooth motion commands are generated based on the different distances, driving the industrial robot 1 to achieve rapid posture approach across six axes. The laser spot is guided into the first PSD5 and simultaneously reflected by the first reflector 9 to the 5mm×5mm effective photosensitive area of ​​the second PSD6, achieving laser spot entry into the window. Gradual speed control is used throughout the process to avoid beam deviation caused by sudden robot movements or jitter.

[0055] In this embodiment, the average time taken for the industrial robot 1 to complete the laser spot entry into the first PSD5 from the initial distance (maximum distance 50mm) and then be reflected by the first reflector 9 to the effective photosensitive area of ​​the second PSD6 is ≤3s.

[0056] If the initial distance is too large and the monocular industrial camera 3 fails to detect the laser spot, the system automatically triggers the fault-tolerant search logic, controlling the industrial robot 1 to perform a slow, equidistant spiral traversal trajectory from the inside out, with the first PSD5 and the second PSD6 as the center, gradually scanning the entire area around the target until the laser spot is successfully detected, and automatically switching to the subsequent alignment process, without any manual intervention throughout the entire process.

[0057] Step 22. The first set of dual PSDs (first PSD5 and second PSD6) is closed-loop locked to the first optical path measurement group to achieve high-precision alignment;

[0058] When the monocular industrial camera 3 visually detects that the laser spot has completely fallen into the effective photosensitive area of ​​the first PSD5 and the second PSD6 reflected by the first reflector 9, the system automatically shuts down the visual position feedback channel and switches to the PSD high-precision closed-loop control mode. At this time, the data acquisition unit 11 acquires the two-dimensional coordinates of the center of the spot projected onto the first PSD5 and the second PSD6, calculates the deviation (spot center deviation) between the center of the spot and the center of the effective photosensitive area of ​​the first PSD5 and the second PSD6, and obtains the two-dimensional deviation vector of the spot.

[0059] In this embodiment, a piecewise variable parameter adaptive PID control algorithm is adopted to adapt to the convergence requirements of different deviation ranges: when the center deviation of the light spot is greater than 2mm, the proportional coefficient is increased and the integral action is weakened to quickly reduce the light spot offset and improve the convergence speed; when the center deviation of the light spot is in the range of 0.5~2mm, the PID parameters are balanced to ensure convergence stability; when the center deviation of the light spot is less than 0.5mm, the proportional coefficient is reduced, the integral fine-tuning is enhanced, and a small derivative is introduced to suppress oscillation and avoid overshoot rebound.

[0060] After iterative closed-loop adjustment, when the center deviation of the laser spot on the first PSD5 and the second PSD6 simultaneously meets the spot plane alignment threshold, that is, when the center deviation of the laser spot stably converges to within 0.1mm of the spot plane alignment threshold, and the spot has no jumping or drifting, the first set of dual PSDs achieves high-precision alignment. The first optical path measurement group is locked and the six joint angle values, end-effector TCP pose, two-dimensional coordinates of the center of the laser spot on the first PSD5 and the second PSD6, and the two-dimensional deviation vector are recorded as the first set of data.

[0061] Step 3. Second projection: Use the second optical path measurement group to achieve joint alignment of vision and PSD, and store the second set of data;

[0062] like Figure 4 As shown, the same laser is kept continuously on, and the industrial robot 1 is controlled to switch from the first preset station to the second preset station, so that the same laser beam is projected a second time onto the third PSD7 and the fourth PSD8. This projection also follows the joint workflow of vision-guided window entry to PSD closed-loop locking, and the specific implementation process is as follows:

[0063] Step 31. Visually guide the second projection from a distance to achieve laser spot entry into the window;

[0064] After the first set of data is locked, the system keeps laser 2 constantly on and reactivates the visual position feedback channel. Based on the preset second preset workstation coordinates, the monocular industrial camera 3 re-acquires images and calculates the distance of the current laser spot relative to the effective photosensitive areas of the third PSD7 and the fourth PSD8. A smooth motion command is generated to drive the industrial robot 1 to move smoothly from the first preset workstation to the second preset workstation, guiding the projection point of the laser beam emitted by laser 2 from the first PSD5 and the second PSD6 areas to the effective photosensitive areas of the third PSD7 and the fourth PSD8, thus achieving laser spot entry into the window.

[0065] During cross-group switching, the monocular industrial camera 3 provides position feedback throughout the process, ensuring that the laser spot is not lost during the workstation switching. If the laser spot is lost during the switching process, the system also triggers the spiral search fault-tolerant logic, controlling the industrial robot 1 to execute an equidistant spiral traversal trajectory from the center of the third PSD7 and the fourth PSD8, until the laser spot is re-identified and imported into the third PSD7 and reflected by the second reflector 10 to the fourth PSD8 window. In this embodiment, the average time for cross-group switching and secondary window entry is ≤4s.

[0066] Step 32. The second set of dual PSDs (third PSD7 and fourth PSD8) is used to lock the second optical path measurement group in a closed loop to achieve high-precision alignment;

[0067] When the monocular industrial camera 3 visually determines that the light spot has entered the effective photosensitive area of ​​the third PSD7 and the fourth PSD8 reflected by the second reflector 10, it closes the visual feedback channel again and switches to the high-precision closed-loop control mode of the third PSD7 and the fourth PSD8. At this time, the data acquisition unit 11 acquires the two-dimensional coordinates of the center of the light spot projected onto the third PSD7 and the fourth PSD8, calculates the deviation (light spot center deviation) between the center of the light spot and the center of the effective photosensitive area of ​​the third PSD7 and the fourth PSD8, and obtains the two-dimensional deviation vector of the light spot. Using the same piecewise variable parameter adaptive PID control algorithm as in step 22, the light spot center deviation of the laser light spot center on the third PSD7 and the fourth PSD8 is simultaneously and stably converged to within the light spot plane alignment threshold, completing the second set of dual PSD high-precision alignment. The second optical path measurement group is locked and the current six joint angle values, end-effector TCP pose, two-dimensional coordinates of the center of the light spot on the third PSD7 and the fourth PSD8, and the two-dimensional deviation vector are recorded as the second set of data.

[0068] Step 4. Fine alignment of dual-optical-path non-planar spatial geometric constraints, such as... Figure 5 As shown and Figure 6 As shown;

[0069] After both the first and second projection alignments are completed, the system initiates four-PSD dual-optical-path non-planar constraint fine alignment control. Based on the robot end effector TCP pose locked during the two projections, the spatial linear equations of the first projection laser path L1 and the second projection laser path L2 are reconstructed respectively.

[0070] The origin of laser 2 during the first projection is set to be in the world coordinate system. The unit direction vector of the beam is During the second projection, the origin of laser 2's emission point in the world coordinate system is... The unit direction vector of the beam is The linear distance between the two optical paths is... for:

[0071] ;

[0072] Real-time calculation of linear distance between two optical paths Distance from theoretical standard skew line distance deviation The fixed mechanical geometry of the dual optical paths in the four PSD calibration device is given in advance, wherein, .

[0073] Based on the spot center deviations of the first PSD5 and the second PSD6 obtained in step 22, the spot center deviations of the third PSD7 and the fourth PSD8 obtained in step 32, and the deviations of the straight-line distance between opposite planes, a multi-objective coupled evaluation function is constructed:

[0074] ;

[0075] In the formula, , These are the two-dimensional deviation vectors of the light spots on the first PSD and the second PSD in the first optical path measurement group, respectively. , These are the two-dimensional deviation vectors of the light spots on the third and fourth PSDs in the second optical path measurement group, respectively. This refers to the deviation in distance between skew lines. , , All are weighted coefficients and satisfy > and > .

[0076] In this embodiment, the geometric constraint weight is increased, that is, by making , Prioritize ensuring the spatial geometric accuracy of the dual optical paths to prevent the problem of single-path alignment being qualified while the overall calibration model fails.

[0077] In this embodiment, minimizing the evaluation function is taken as the optimization objective. A gradient descent iterative optimization algorithm is used in the robot joint space to calculate the six-degree-of-freedom pose correction of the industrial robot, and the robot's end-effector pose is fine-tuned cycle by cycle. Since the first and second projection poses jointly determine the skew lines, i.e., the spatial positions of the first projection laser path L1 and the second projection laser path L2, they are highly coupled under spatial geometric constraints. Therefore, adjusting the first projection pose to improve the skew line distance deviation... Therefore, the second projection pose adjustment must be triggered synchronously; otherwise, unilaterally adjusting the first projection pose will inevitably destroy the alignment conditions of the first set of light spot planes that have already been met. Similarly, unilaterally adjusting the second projection pose will also destroy the alignment conditions of the second set of light spot planes. This bidirectional linkage compensation mechanism driven by the deviation of the linear distance between opposite planes is a unique and necessary technical means in the single-laser time-division projection architecture to ensure that the deviation of the two sets of light spot planes does not exceed the tolerance due to spatial constraint adjustment. It is by no means a simple superposition of two independent alignments.

[0078] This implementation performs the aforementioned bidirectional linkage compensation in each gradient descent iteration to ensure that the center deviation of the light spot on the four PSDs remains within 0.1mm during the convergence of the non-planar straight-line distance. During iteration, dual composite constraint thresholds are set in real time: a light spot plane alignment threshold of 0.1mm and a non-planar distance deviation threshold of 0.2mm. The dual composite constraints are automatically verified after each iteration. Only when the center deviation of the light spot (all four PSDs simultaneously satisfy ≤0.1mm) and the non-planar straight-line distance deviation (…) are both satisfied… When both conditions are met, the precision alignment is considered complete.

[0079] Step 5. Fully automated multi-station iterative data acquisition and calibration parameter solution; complete robot calibration;

[0080] In this embodiment, after a single dual-path synchronous precision alignment and dual composite constraint verification pass, the system automatically triggers a data storage command to record all calibration data at the current moment with high precision. Specifically, this includes: the first set of data: the angle values ​​of the six joints of the industrial robot 1, the end effector TCP pose, and the two-dimensional coordinates and two-dimensional deviation vectors of the light spot centers on the first PSD5 and the second PSD6; the second set of data: the angle values ​​of the six joints of the industrial robot 1, the end effector TCP pose, and the two-dimensional coordinates and two-dimensional deviation vectors of the light spot centers on the third PSD7 and the fourth PSD8; and the linear distance between the two optical paths. skew line distance deviation Ambient temperature and equipment operating parameters;

[0081] After completing a set of data acquisitions, the system automatically calls the preset calibration pose library (30 sets of differentiated calibration poses are preset in this embodiment) without manual intervention, smoothly switches to the next set of industrial robot poses, and repeats steps 2 to 5 until all 30 sets of data acquisitions are completed. Based on the 30 sets of valid samples, the system constructs a set of overdetermined kinematic equations for the robot. Combined with the geometric constraints of the dual-optical-path skew space, the system uses a weighted least squares iterative algorithm to identify a complete set of calibration parameters, including robot link parameters, joint zero-bias error, laser end-effector installation offset error, and laser beam spatial skew error, and completes the robot calibration.

[0082] Specific Implementation Method Two: This implementation method is a verification example of the vision-PSD joint guidance robot calibration automatic alignment method for single laser time-division projection described in Specific Implementation Method One:

[0083] To fully verify the advancement, stability and effectiveness of the present invention compared with the prior art, two sets of existing mainstream calibration schemes were set up as comparative examples under the same hardware platform, robot working conditions, ambient lighting and calibration constraints. Parallel comparative experiments were carried out with the method of the present invention, and the number of test samples, calibration pose and judgment threshold were kept uniform.

[0084] Existing Solution 1: Manual teaching and fine-tuning calibration;

[0085] The traditional calibration method, commonly used in the industry, involves manual visual alignment, manual teaching and posture adjustment, manual convergence judgment, and manual data recording. Experimental results show that the entire calibration process requires continuous intervention from professional operators, taking over 45 minutes to complete, resulting in extremely low efficiency. Affected by human visual discrimination ability, operator feel, and subjective judgment errors, the repeatability of the light spot alignment is poor. Multiple tests showed that the light spot alignment error ranged from 0.5 to 1.0 mm, exhibiting poor steady-state accuracy and large dispersion. Furthermore, manual methods cannot accurately quantify the geometric constraints of the dual-optical-path non-planar geometry, relying solely on experience to judge light spot alignment. This leads to poor geometric consistency among calibration samples, high data noise, and ultimately, large residuals in the identified robot kinematic parameters and low calibration stability.

[0086] Existing Solution 2: Single-shot hybrid servo alignment of vision and PSD;

[0087] The existing technology, namely the traditional vision and PSD hybrid servo method, is a single-direction switching mode from coarse vision guidance to fine PSD locking. It can only achieve planar alignment of a single spot to a single PSD target under single projection conditions, without a time-division dual-optical-path alignment mechanism or fine alignment with dual-optical-path non-plane spatial geometric constraints.

[0088] Experimental results show that this method can achieve large field-of-view spot search and preliminary alignment, and can complete single PSD spot windowing and two-dimensional plane approximation. However, limited by visual imaging resolution and depth-of-field error, the near-range steady-state alignment error is relatively large (approximately 0.3~0.5mm), and micrometer-level precise convergence cannot be achieved. The most critical drawback is that this type of scheme only performs one projection and one locking, completely ignoring the spatial fusion of two time-division projection data. It cannot construct dual-path non-plane geometric constraints, resulting in a large number of pseudo-qualified samples where the single-path spot alignment is qualified, but the dual-path spatial configuration does not meet the modeling conditions. This leads to missing constraints in the four-PSD calibration equation system, ill-conditioned matrices, and model failure, completely failing to meet the modeling requirements of four-PSD non-plane coupling precise calibration. The comprehensive comparison conclusions between the method of this invention and the above-mentioned prior art are shown in Table 1.

[0089] Table 1

[0090]

[0091] As shown in Table 1, compared to the existing solution 1, the method of this invention achieves fully unmanned automated operation, reducing calibration time to 12 minutes, increasing efficiency by 3.75 times, and improving the steady-state alignment accuracy of the light spot by an order of magnitude. Compared to the existing solution 2, the method of this invention balances large-scale search capability with micron-level convergence accuracy through a vision-PSD-vision-PSD multi-round joint guidance mechanism. Simultaneously, it innovatively employs a dual-optical-path heterogeneous coupling multi-target constraint mechanism, fundamentally solving the core problems of calibration model failure and invalid sample constraints. All collected samples achieve a 100% constraint compliance rate, exhibiting good data consistency, low noise, and strong robustness. In terms of calibration accuracy, automation level, adaptability to operating conditions, and modeling reliability, it significantly outperforms existing solutions.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. An automatic alignment method for robot calibration based on vision and PSD joint guidance, characterized by: This method is implemented by the following steps: Step 1: Construct a calibration system to achieve pre-calibration of multiple coordinate systems and build a pre-calibrated four-coordinate system mapping model; Step 2: First projection, using the first optical path measurement group to achieve joint alignment of vision and PSD; The robot is controlled to move to the first preset station. The laser spot is guided into the window from a distance by vision. Then, the first set of double PSD precision alignment is completed through PSD closed-loop control, and the first set of data is stored. Step 3: Second projection, using the second optical path measurement group to achieve joint alignment of vision and PSD, and storing the second set of data; The laser continues to work, controlling the robot to switch to the second preset station. After vision cross-group navigation, the second set of dual PSD precision alignment is completed again through PSD closed-loop control, and the second set of data is stored. Step 4: Based on the first set of data obtained in Step 2 and the second set of data obtained in Step 3, perform precise alignment using dual-optical-path non-plane spatial geometric constraints. Based on two projections, a non-plane laser optical path is established, the straight-line distance between the two optical paths is calculated, a multi-target coupled evaluation function is constructed, and the dual optical paths are optimized using a gradient descent iterative algorithm. A dual composite constraint threshold is set: a spot plane alignment threshold and a non-plane distance deviation threshold. The dual composite constraint is automatically verified after each iteration. When the center deviation of the two spot paths and the non-plane straight-line distance deviation both meet the standards, the fine alignment is determined to be completed. Step 5: Automatically collect multiple sets of posture data to complete the robot calibration and automatic alignment.

2. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 1, characterized in that: The calibration system includes a robot, a laser, a monocular industrial camera, a four-PSD calibration device, and a data acquisition unit; The laser is fixed to the end of the robot, and the monocular industrial camera is fixed directly above the center of the four-PSD calibration device; the four-PSD calibration device includes a first PSD, a second PSD, a third PSD, and a fourth PSD; and a first reflector and a second reflector placed above the working surface of the four-PSD calibration device; The first PSD, the first reflector, and the second PSD constitute the first optical path measurement group, and the third PSD, the second reflector, and the fourth PSD constitute the second optical path measurement group. The data acquisition unit is used to synchronously acquire visual images, PSD analog signals, and robot joint data.

3. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 2, characterized in that: In step one, the multi-coordinate system pre-calibration process is as follows: Zhang's calibration method was used to complete the intrinsic parameter calibration and image distortion correction of a monocular industrial camera. Based on multi-point pose matching of the calibration board, a fixed extrinsic homogeneous transformation matrix of the camera coordinate system relative to the global coordinate system of the four-PSD calibration device is obtained by fitting, and the mapping relationship between the camera coordinate system and the global coordinate system of the four-PSD calibration device is established. The robot's motion is controlled to obtain multiple sets of spatial discrete teaching poses. Combined with the coordinates of the laser spot on the four PSD calibration device and the relationship between the robot and the laser installation structure, the mapping relationship between the robot coordinate system and the PSD measurement coordinate system is obtained by iterative fitting using the least squares algorithm. Based on the inherent mapping relationship between the global coordinate system of the four-PSD calibration device and the PSD measurement coordinate system determined by the spatial structure, a pre-calibrated four-coordinate system mapping model is established, consisting of the robot coordinate system, the global coordinate system of the four-PSD calibration device, the PSD measurement coordinate system, and the camera coordinate system.

4. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 3, characterized in that: In step two, the first projection is guided by visual distance to achieve laser spot entry into the window and the first set of dual PSD closed-loop locking of the first optical path measurement group to achieve high-precision alignment; Visual long-distance guidance for the first projection enables the laser spot to enter the window; The robot is controlled to move from its current arbitrary pose to the first preset workstation, so that the laser beam emitted by the laser is projected onto the first PSD and reflected by the first reflector to the second PSD. The monocular industrial camera acquires the image. Based on the pre-calibrated four-coordinate system mapping model obtained in step one, the distance of the current laser spot relative to the effective photosensitive area of ​​the first PSD and the second PSD is calculated. The robot is then driven to guide the laser spot into the effective photosensitive area of ​​the first PSD and the second PSD, thus realizing the laser spot entering the window. The first group of dual PSD closed-loop locks the first optical path measurement group to achieve high-precision alignment; The visual position feedback channel is turned off, and the system switches to the PSD high-precision closed-loop control mode. The data acquisition unit acquires the two-dimensional coordinates of the light spot center projected onto the first PSD and the second PSD, calculates the deviation of the light spot center from the center of the effective photosensitive area of ​​the first PSD and the second PSD, i.e., the light spot center deviation, and obtains the two-dimensional deviation vector of the light spot. When the deviation of the laser spot center on the first PSD and the second PSD simultaneously meets the spot plane alignment threshold, the first set of dual PSDs is aligned with high precision. The six joint angle values ​​of the robot, the end-effector TCP pose, the two-dimensional coordinates of the spot center on the first PSD and the second PSD, and the two-dimensional deviation vector are recorded as the first set of data.

5. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 4, characterized in that: In step three, high-precision alignment is achieved by visually guiding the second projected laser beam into the window and locking the second optical path measurement group with a second set of dual PSD closed loops. Visual guidance from a distance enables a second projection, allowing the laser beam to enter the window; The visual position feedback channel is activated, and the robot is switched from the first preset station to the second preset station. The laser beam is projected onto the third PSD and reflected by the second mirror to the fourth PSD. The monocular industrial camera re-acquires the image. Based on the pre-calibrated four-coordinate system mapping model obtained in step one, the distance of the current laser spot relative to the effective photosensitive areas of the third and fourth PSDs is calculated. The robot is then driven to guide the laser spot to the effective photosensitive areas of the third and fourth PSDs, thus realizing the laser spot entering the window. The second group of dual PSD closed-loop locks the second optical path measurement group to achieve high-precision alignment. The visual position feedback channel is turned off again, and the PSD high-precision closed-loop control mode is switched. The two-dimensional coordinates of the light spot center projected onto the third PSD and the fourth PSD are collected by the data acquisition unit. The deviation of the light spot center from the effective photosensitive area center of the third PSD and the fourth PSD is calculated respectively, and the two-dimensional deviation vector of the light spot is obtained. When the laser spot center deviation on the third PSD and the fourth PSD simultaneously meets the spot plane alignment threshold, the second set of dual PSDs is aligned with high precision. The current six joint angle values ​​of the robot, the end-effector TCP pose, the two-dimensional coordinates of the spot center on the third PSD and the fourth PSD, and the two-dimensional deviation vector are recorded as the second set of data.

6. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 5, characterized in that: A piecewise variable parameter adaptive PID control algorithm is adopted to determine whether the spot center deviation on the first PSD and the second PSD simultaneously meets the spot plane alignment threshold, and to determine whether the spot center deviation on the third PSD and the fourth PSD simultaneously meets the spot plane alignment threshold; the spot plane alignment threshold is 0.1mm.

7. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 6, characterized in that: During the first projection, if the initial distance exceeds the maximum distance and the monocular industrial camera fails to identify the laser spot, the system automatically triggers the fault-tolerant search logic, controlling the robot to perform a slow, equidistant spiral traversal trajectory from the inside out, with the center of the first PSD and the center of the second PSD as the center, until the laser spot is successfully identified. During the cross-group switching process, the monocular industrial camera provides position feedback through vision throughout the process. If the laser spot is lost during the switching process, the system triggers the spiral search fault-tolerant logic, controlling the robot to execute an equidistant spiral traversal trajectory from the inside out with the center of the third PSD and the fourth PSD as the center, until the laser spot is re-identified and imported into the third PSD and reflected by the second reflector to the fourth PSD window.

8. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 7, characterized in that: In step four, based on the end TCP poses recorded by the first and second projections, the spatial linear equations of the first projection laser path L1 and the second projection laser path L2 are reconstructed respectively. The origin of the laser emission point in the world coordinate system is set to [value] during the first projection. The unit direction vector of the beam is During the second projection, the laser's emission origin in the world coordinate system is... The unit direction vector of the beam is The straight-line distance between the two optical paths is as follows: for: ； Construct a multi-objective coupled evaluation function: ； In the formula, , These are the two-dimensional deviation vectors of the light spots on the first PSD and the second PSD, respectively; , These are the two-dimensional deviation vectors of the light spots on the third and fourth PSDs, respectively; This refers to the deviation in distance between skew lines. , , All are weighting coefficients; With minimizing the multi-objective coupled evaluation function as the optimization objective, the gradient descent iterative optimization algorithm is used to calculate the correction amount of the robot's six-axis joints and adjust the robot's end-effector TCP pose cycle by cycle. During the iterative optimization process, when the center deviation of the two light spots and the deviation of the distance between the opposite planes reach the target at the same time, it is determined that the fine alignment is completed.

9. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 8, characterized in that: The threshold for the non-plane distance deviation is 0.2 mm.

10. The automatic alignment method for robot calibration based on vision and PSD joint guidance according to claim 9, characterized in that: In step five, all calibration data are recorded, including the first set of data, the second set of data, the distance between the two optical paths in opposite directions, the deviation of the distance between the two optical paths, the ambient temperature, and the equipment operating parameters. After completing the acquisition of one set of data, the preset calibration posture library is automatically called, and the robot smoothly switches to the next set of differentiated robot postures. Steps two to five are repeated to collect multiple sets of data, construct the robot's kinematic overdetermined equations, and combine the geometric constraints of the two optical paths in opposite directions. The weighted least squares iterative algorithm is used to identify the robot's link parameters, joint zero-bias error, laser end effector installation offset error, and laser beam spatial skew error in batches, thus completing the robot calibration.

Images