Hand-eye calibration method suitable for five-axis series-parallel mechanical arm

Abstract

The invention discloses a hand-eye calibration method suitable for a five-axis series-parallel mechanical arm, which comprises the following steps of: analyzing coordinate conversion error distribution characteristics according to an initial hand-eye calibration matrix calculation result, and separating a total error into two components, namely a translation error and a rotation error through an error vector decomposition algorithm, obtaining distribution rules and variation trends of the errors in different spatial directions; establishing a calibration precision evaluation system according to the verification result, calculating the repeated positioning precision and the absolute positioning precision of the system by counting and analyzing the position deviation data of multiple tests, and obtaining the system performance index in the current calibration state; a self-adaptive updating mechanism is established for system performance indexes, local updating of calibration parameters is automatically triggered when environment temperature change or small deformation of a mechanical structure is detected, and a compensation model is adjusted in an incremental learning mode to avoid comprehensive recalibration.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of robot vision systems, and in particular to a hand-eye calibration method suitable for a 5-axis hybrid mechanical arm. BACKGROUND

[0002] In a robot vision system, hand-eye calibration is a key step for determining the relationship between the camera coordinate system and the robot end coordinate system. Traditional hand-eye calibration methods for serial 6-axis mechanical arms usually rely on accurate motion of the robot end effector, and through a series of known robot poses and image data collected by the camera, a mathematical optimization algorithm is used to solve the transformation matrix between the camera and the robot end; as a core technology of modern intelligent manufacturing, a robot vision system plays an irreplaceable role in key applications such as industrial automation, precision assembly and quality detection, and through the combination of robot operation capability and visual perception capability, high-precision automated operation is realized, which is an important driving force for the intelligent transformation of the manufacturing industry.

[0003] Current mainstream hand-eye calibration methods mainly rely on complex mathematical modeling and a large amount of calibration data collection, and these methods often face problems such as unstable calibration accuracy, harsh environmental requirements, and complicated calibration processes in actual applications. Traditional calibration methods usually require accurate initial parameter settings and high professional skills of the operator, which is difficult to meet the requirements of rapid deployment and high precision in industrial field, and the fundamental challenge faced by the robot vision system is that the accurate conversion relationship between the camera coordinate system and the robot coordinate system is difficult to establish.

[0004] Due to factors such as mechanical assembly errors and thermal deformation, the initial hand-eye calibration matrix obtained by measuring tools often has large deviations, and this initial error will directly affect the accuracy of all subsequent visual guidance operations, and the existence of initial calibration error further leads to the problem of cumulative error in the coordinate conversion process. When the target point coordinates detected by the camera are converted to the robot base coordinate system, the conversion error will be amplified with the increase of the operation distance and angle, and the cumulative error problem makes the robot unable to accurately reach the target position specified by the vision system, which seriously restricts the operation accuracy and reliability of the entire system, especially in precision assembly and detection tasks that require high-precision positioning, such error is often unacceptable.

[0005] Therefore, how to realize high-precision registration between the camera coordinate system and the robot coordinate system under the condition of existing initial measurement error through effective data collection and processing methods, and eliminate the cumulative error in the coordinate conversion process, has become a key problem for the robot vision system to realize high-precision operation. SUMMARY

[0006] The application provides a hand-eye calibration method suitable for a 5-axis hybrid mechanical arm, mainly comprising the following steps: Step 1), a multi-point sampling calibration algorithm is used to obtain coordinate data of a robot end effector at different positions and attitudes, corresponding image feature point coordinates are synchronously collected by a camera, an initial hand-eye calibration matrix is established, and if the number of collected points is less than a preset threshold, the sampling points are continuously increased until the matrix calculation accuracy requirement is met; Step 2), the coordinate conversion error distribution characteristics are analyzed according to the initial hand-eye calibration matrix calculation result, the total error is separated into two components of translational error and rotational error through an error vector decomposition algorithm, and the distribution law and variation trend of the error in different spatial directions are obtained; Step 3), a dynamic compensation model is established according to the error variation trend, a least square method is used to fit the functional relationship between the error and the robot operation distance and angle, if the fitting residual exceeds the preset accuracy range, the fitting order is adjusted to recalculate until the compensation function meeting the accuracy requirement is obtained; Step 4), the original hand-eye calibration matrix is corrected in real time through the correction parameters calculated by the compensation function, the corresponding error compensation amount is determined according to the current robot position and the target point distance, and the compensation amount is superimposed into the coordinate conversion process to obtain the corrected target coordinates; Step 5), an iterative optimization algorithm is used to verify the corrected coordinate conversion result, the deviation data of the actual reaching position and the theoretical target position are obtained by the robot executing a standard test action, if the deviation exceeds the allowable range, the compensation model parameters are returned to step three for readjustment; Step 6), a calibration accuracy evaluation system is established according to the verification result, the repeatability and absolute positioning accuracy of the system are calculated by statistically analyzing the position deviation data of multiple tests, and the system performance index under the current calibration state is obtained; Step 7), an adaptive updating mechanism is established for the system performance index, when the environmental temperature change or the slight deformation of the mechanical structure is detected, the local update of the calibration parameters is automatically triggered, and the compensation model is adjusted through the incremental learning method to avoid full recalibration; Step 8), the coordinate conversion accuracy change in the robot operation process is monitored in real time, the positioning error trend of the recent operations is analyzed by using a sliding window algorithm, if the error presents a systematic growth, the automatic optimization process of the calibration parameters is started to ensure the long-term stable operation of the system. Beneficial effects

[0007] The application establishes an initial hand-eye calibration matrix by collecting coordinate data and corresponding image feature point coordinates of a robot end effector at different positions, analyzes the distribution characteristics of the coordinate conversion error and establishes a dynamic compensation model, uses the least square method to fit the functional relationship between the error and the robot operation parameters, corrects the original calibration matrix in real time through the compensation function, verifies the correction result by using an iterative optimization algorithm, and establishes a calibration accuracy evaluation system and an adaptive updating mechanism, which can automatically trigger local parameter updating according to environmental changes and start the automatic optimization process by monitoring the accuracy change in real time. The method significantly improves the accuracy and stability of the robot hand-eye calibration and realizes long-term reliable operation. DETAILED DESCRIPTION

[0008] The technical solutions of the application will be described below in conjunction with the embodiments. Obviously, the described embodiments are only some of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0009] A hand-eye calibration method suitable for a 5-axis hybrid mechanical arm can specifically include the following steps: Step 1), a multi-point sampling calibration algorithm is used to obtain coordinate data of a robot end effector at different positions and attitudes, corresponding image feature point coordinates are collected by a camera in synchronization, an initial hand-eye calibration matrix is established, and if the number of collected points is less than a preset threshold, the sampling points are continuously increased until the matrix calculation accuracy requirement is met.

[0010] Specifically, coordinate data of a robot end effector at different positions and attitudes are obtained by using a multi-point sampling calibration algorithm, corresponding image feature point coordinates are collected by a camera in synchronization to obtain an initial data set, the initial data set is processed by using the least square method, a hand-eye calibration matrix is constructed, and initial matrix parameters are determined; if the calculation error of the initial matrix exceeds a preset threshold, the sampling point positions are adjusted by using an iterative optimization algorithm to obtain new coordinate data and feature point coordinates, the hand-eye calibration matrix is updated according to the newly obtained coordinate data and feature point coordinates, an optimized matrix is obtained, the optimized matrix is decomposed by using a singular value decomposition algorithm, main feature vectors are extracted, and it is judged whether the matrix accuracy meets the requirement; if the matrix accuracy is lower than the preset threshold, the sampling point distribution strategy is adjusted according to the distribution characteristics of the feature vectors, supplementary coordinate data are obtained, the hand-eye calibration matrix is reconstructed by using the supplementary coordinate data, and a final calibration result is obtained.

[0011] In a possible implementation of the present application, the robot end effector coordinate data is obtained by a multi-point sampling calibration algorithm, the core of which is to ensure the diversity of the sampling points covering the workspace. For example, in an industrial assembly scenario, the robot end effector can be selected at 10 different positions in the workspace, such as spatial coordinates (100, 200, 300) mm, (150, 250, 350) mm, etc., combined with different attitude angles, such as rotating 0°, 45°, 90° around the Z axis, to obtain three-dimensional coordinate data. At the same time, the camera synchronously collects image feature points, such as circular marker points on the workpiece, and records pixel coordinates such as (320, 240), (400, 300). These data constitute an initial data set, providing a basis for subsequent calibration. The diversified sampling point distribution can improve the robustness of the calibration matrix and reduce the deviation of single region sampling. Specifically, the least squares method is used to process the initial data set to construct a hand-eye calibration matrix. For example, based on 10 sets of coordinate data and corresponding feature points, the transformation matrix of the robot end to the camera coordinate system is calculated. The initial matrix parameters may include a rotation component and a translation component, such as a rotation matrix R and a displacement vector T. Error evaluation is completed by comparing the deviation of the predicted feature point position and the actual position. If the average error exceeds the preset threshold of 0.5 mm, optimization is required. The least squares method improves the initial accuracy of the matrix by optimizing the sum of squared residuals, laying a foundation for subsequent iterations.

[0012] In an embodiment of the present application, if the error of the initial matrix exceeds the standard, the iterative optimization algorithm can adjust the sampling point position. For example, the initial sampling points are concentrated in the central region of the workspace, and the error may be high due to insufficient boundary data. By optimizing with a genetic algorithm, new boundary points such as (50, 50, 50) mm and (200, 300, 400) mm are selected to obtain new coordinate and feature point data. This strategy can effectively cover the boundary of the workspace and improve the adaptability of the matrix to complex poses. The error of the updated hand-eye calibration matrix can be reduced to 0.3 mm, significantly improving the calibration accuracy. For example, the singular value decomposition algorithm decomposes the optimized matrix to extract the main eigenvector to evaluate the matrix accuracy. After decomposition, if the distribution of the main eigenvector shows that the matrix is insufficient to describe the transformation in some directions, such as the eigenvalue of the Z-axis direction being small, it indicates that the sampling points are insufficient in the Z-axis direction. The adjustment strategy can increase the sampling points in the Z-axis direction, such as (100, 200, 400) mm. After supplementing the data, the matrix is reconstructed. This method accurately locates the problem by analyzing the eigenvector, ensuring that the matrix accuracy meets the preset threshold of 0.2 mm.

[0013] In a possible implementation of the present application, the hand-eye calibration matrix is reconstructed after supplementing the coordinate data, and the final calibration result is obtained. For example, after adding 5 groups of sampling points, the translation error of the updated matrix is reduced to 0.15 mm, and the rotation error is reduced to 0.1°. The final calibration result is applied to the robot vision guidance task, such as precise grabbing of workpieces. The positioning accuracy is improved by 30%, which significantly reduces the positioning failure rate in production. The strategy of supplementing data ensures full coverage of the matrix to the workspace, enhances the stability and reliability of the system in complex environments. It should be noted that the above method forms a closed-loop calibration process through multi-point sampling, iterative optimization and feature vector analysis. Each step supports each other to ensure that the accuracy of the calibration matrix is gradually improved, and finally the high-precision hand-eye coordination is realized. The advantage of this method is strong adaptability, which can cope with different robots and camera configurations, and is widely applicable to industrial automation scenarios.

[0014] Step 2), analyze the coordinate conversion error distribution characteristics according to the calculation result of the initial hand-eye calibration matrix, separate the total error into translation error and rotation error two components through error vector decomposition algorithm, and obtain the distribution law and change trend of the error in different spatial directions.

[0015] The coordinate conversion error data is obtained through the initial hand-eye calibration matrix, the translation error and the rotation error are separated by using the error vector decomposition algorithm, the error components are obtained, the distribution characteristics of the translation error and the rotation error in the spatial coordinate system are extracted from the error components, the error distribution characteristics are determined, if the error distribution characteristics exceed the preset threshold, the spatial direction decomposition is performed on the translation error and the rotation error, the spatial direction distribution is obtained, the error vector modulus in each direction is calculated according to the spatial direction distribution, the distribution law extraction result is obtained, the dimensionality reduction processing is performed on the distribution law extraction result by using the principal component analysis algorithm, the change trend analysis data is obtained, the error change curve is fitted through the change trend analysis data, the change trend of the error in the spatial direction is determined, the key feature points are extracted from the change trend, and the feature description of the spatial direction error distribution is generated.

[0016] Specifically, in the hand-eye calibration scenario, obtaining coordinate transformation error data through the initial hand-eye calibration matrix is the basis for achieving high-precision robot operation. For example, in an industrial robot assembly scenario, a mechanical arm locates a target object through a vision system, and a calibration matrix relates the camera coordinate system to the mechanical arm coordinate system. Coordinate transformation error data is usually obtained by measuring the deviation between the actual position and the theoretical position multiple times. Assuming that the mechanical arm locates a workpiece in space, the theoretical coordinates are (100, 200, 300) mm, and the actual measured coordinates are (102, 199, 301) mm, the error vector is (2, -1, 1) mm. This process can collect multiple sets of data through high-precision sensors to generate an error dataset. The key step is to separate the translation error and the rotation error using an error vector decomposition algorithm. The translation error refers to the positional deviation, and the rotation error refers to the attitude deviation. For example, in the above example, the error vector (2, -1, 1) mm can be directly considered as the translation error, and the rotation error can be calculated through the Euler angle deviation of the attitude matrix. Assuming that the calibration matrix indicates that the mechanical arm has a 0.5-degree rotation deviation around the Z-axis and a 0.2-degree deviation around the X and Y axes, the error components can be separated through the vector decomposition algorithm. This separation helps to analyze the impact of translation and rotation on system accuracy and improves calibration optimization efficiency. Extracting spatial distribution features from error components requires analyzing the error in the three-dimensional coordinate system. Specifically, the mean and variance of the translation error on the X, Y, and Z axes can be calculated through statistical analysis. For example, after collecting 100 sets of data, it is found that the mean of the X-axis translation error is 2 mm, and the variance is 0.5 mm². The mean of the Y-axis is -1 mm, and the variance is 0.3 mm², indicating that the X-axis error distribution is more dispersed. Similarly, the rotation error can be analyzed for deviation distribution around each axis. If the error distribution characteristics exceed the preset threshold (e.g., the mean translation error exceeds 3 mm or the rotation error exceeds 1 degree), the spatial direction distribution needs to be further decomposed. For example, the translation error is projected along the X, Y, and Z axes to obtain the error vector modulus in each direction, and the distribution rule is analyzed to determine the main error source. Principal component analysis algorithm is used to reduce the dimensionality of the distribution rule results, which can effectively extract the change trend. For example, through principal component analysis, it is found that the X-axis translation error and the Z-axis rotation error contribute 80% of the error change, indicating that the optimization focus should be on these two directions. The reduced data can be used to fit the error change curve. For example, the fitting result shows that the X-axis error increases linearly with the workpiece position, with a slope of 0.02 mm / mm. This curve reveals the error variation rule with spatial position, which helps to predict error behavior. The last step is to extract key feature points from the change trend to generate error distribution feature descriptions.For example, the feature points can include error peak points, inflection points, etc. Assuming that the fitting curve shows that the X-axis error reaches a peak value of 3.5 mm at the workpiece position (150, 200, 300) mm, this feature point can be used to optimize the calibration algorithm, adjust the mechanical arm motion path, and reduce the error at a specific position. This method significantly improves the positioning accuracy of the robot, reduces the assembly failure rate. It should be noted that the above process gradually focuses on the key sources of error by decomposition, analysis, and dimensionality reduction, optimizes the calibration matrix, and ensures high-precision operation. Each step supports each other to form a complete error analysis and optimization chain, significantly improving system stability and production efficiency.

[0017] Step 3), a dynamic compensation model is established for error variation trend, a least squares method is used to fit the functional relationship between error and robot operation distance and angle, if the fitting residual exceeds the preset accuracy range, the fitting order is adjusted to recalculate until the compensation function that meets the accuracy requirement is obtained.

[0018] The robot operation distance and angle data are obtained, real-time operation parameters are collected using sensors, noise is removed through data preprocessing to obtain a standardized operation data set, an initial fitting function of operation distance, angle and error is constructed using a least squares method based on the standardized operation data set, and initial fitting parameters are determined. For the initial fitting function, the fitting residual value is calculated, and whether the accuracy requirement is met is determined by comparing the residual with the preset accuracy threshold. If the residual exceeds the preset accuracy threshold, the fitting order is adjusted, the least squares method is used to calculate the fitting function again to obtain updated fitting parameters. According to the updated fitting parameters, an error dynamic compensation model is constructed to generate a compensation function. Through the compensation function, the real-time operation distance and angle data are error corrected to obtain corrected operation parameters. The corrected operation parameters are used to update the robot control instructions to obtain the optimized motion trajectory.

[0019] For example, in the process of collecting distance and angle data for robot operation, sensors are critical devices, and laser radars or ultrasonic sensors are often used to obtain high-precision real-time data; assuming that in the scenario of industrial robots grabbing workpieces, the laser radar collects distance and angle data at a frequency of 100 times per second, obtaining a raw data set containing noise. In the data preprocessing stage, outliers are removed by median filtering, for example, removing mutations that are 5% outside the normal range of distance data, obtaining a smooth standardized operation data set. This preprocessing ensures the reliability of the data for subsequent analysis, avoiding noise interference with the fitting results. Specifically, based on the standardized data set, the least squares method is used to establish a relationship model between distance, angle and error when constructing the initial fitting function, for example, in the movement of a robot arm, assuming that the operation distance range is 0.5 meters to 2 meters and the angle range is -45 degrees to 45 degrees, a quadratic function is fitted by the least squares method to preliminarily determine parameters such as distance coefficient and angle bias. In the residual error analysis stage, the difference between the predicted value and the actual value of the fitting function is calculated, assuming that the mean residual error is 0.02 meters and the standard deviation is 0.01 meters, and if the preset precision threshold is 0.015 meters, the residual error is out of tolerance. At this time, the fitting function is adjusted to a cubic function, and the mean residual error is reduced to 0.01 meters after re-computation, meeting the accuracy requirement. In an embodiment, a dynamic error compensation model is constructed based on the updated fitting parameters. For example, by analyzing the residual error distribution, it is found that the error increases nonlinearly when the distance is far, and a compensation function can be designed to dynamically adjust the control parameters. Assuming that the error is 0.03 meters at a distance of 1.8 meters, the compensation function corrects this error to within 0.005 meters. The corrected parameters are directly used to update the robot control instructions, for example, adjusting the joint angles of the robotic arm to optimize the motion trajectory, so that the positioning accuracy of grabbing workpieces is improved to 0.002 meters. This dynamic compensation significantly improves the stability of robot operation. For example, in the scenario of updating the motion trajectory, the corrected parameters can be used to adjust the path planning of the robot end effector. Assuming that the robot needs to complete a circular trajectory motion in a plane, the original trajectory has a radius deviation of 0.05 meters due to error, and the corrected deviation is reduced to 0.01 meters. This optimized trajectory ensures the repeatability of workpiece grabbing and improves production efficiency. It should be noted that the real-time nature of error correction depends on the high-frequency acquisition and fast computing power of the sensor, which is particularly important in high-precision assembly tasks. Specifically, the advantage of the above method is that it dynamically adapts to different operation scenarios through data-driven methods, for example, in different workpiece sizes or weights, the error distribution may change, and the dynamic compensation model can adjust the parameters according to real-time data to maintain high precision. The flexibility of this method makes it suitable for a variety of industrial robot tasks, such as welding, assembly or handling, significantly improving operation consistency and reliability.

[0020] Step 4), the original hand-eye calibration matrix is corrected in real time by the correction parameter calculated by the error compensation function, the corresponding error compensation amount is determined according to the current robot position and the target point distance, and the compensation amount is superimposed into the coordinate conversion process to obtain the corrected target coordinates.

[0021] Specifically, the correction parameter is calculated by the error compensation function, and the position error is obtained from the current position of the robot and the distance of the target point. If the position error exceeds the preset threshold, the correction parameter is generated according to the error compensation function. The hand-eye calibration matrix is adjusted in real time by using the correction parameter to obtain the adjusted calibration matrix. The coordinate system transformation is performed according to the adjusted calibration matrix to obtain the initial conversion coordinates. The error compensation amount is calculated from the initial conversion coordinates and the distance of the target point. If the error compensation amount meets the dynamic adjustment condition, the compensation vector is generated. The initial conversion coordinates are superimposed by the compensation vector to obtain the corrected target coordinates. The robot motion control parameters are updated by using the corrected target coordinates, the control instructions are generated, the robot motion is driven according to the control instructions, the real-time position feedback is obtained, and it is judged whether the position error meets the accuracy requirement.

[0022] For example, in the robot hand-eye calibration process, when calculating the correction parameter through the error compensation function, the distance between the current position and the target point can be obtained based on the sensor data first. Assuming that the current coordinates of the robot are (100, 200, 50) millimeters, and the target point coordinates are (105, 205, 55) millimeters, the position error is calculated as (5, 5, 5) millimeters. If the preset threshold is 3 millimeters, the error exceeds the threshold, and the error compensation function needs to be triggered. The error compensation function can generate correction parameters, such as an adjustment coefficient of 0.8, based on historical calibration data and a linear interpolation method, to generate a correction vector (4, 4, 4) millimeters. This method ensures that the correction parameters are consistent with the actual error trend through historical data analysis. In one possible implementation, when adjusting the hand-eye calibration matrix, the matrix parameters can be updated using the correction parameters. Assuming that the original calibration matrix is a 4×4 homogeneous transformation matrix containing rotation and translation components. The translation vector is updated from (10, 20, 30) millimeters to (14, 24, 34) millimeters by adjusting the translation component through the correction parameter. The adjusted matrix can more accurately describe the relative position relationship between the camera and the robot end effector, thereby improving the accuracy of coordinate transformation. Specifically, when performing coordinate system transformation, the initial conversion coordinates can be calculated through the adjusted calibration matrix. For example, the coordinates of the target point in the camera coordinate system are (50, 60, 70) millimeters, and the initial conversion coordinates in the robot coordinate system are (48, 58, 68) millimeters after transformation. At this time, the error compensation amount is calculated by comparing the distance with the target point, and the compensation amount is assumed to be (2, 2, 2) millimeters. If the dynamic adjustment condition requires that the compensation amount be less than 5 millimeters, a compensation vector (2, 2, 2) millimeters is generated and added to the initial conversion coordinates to obtain the corrected target coordinates (50, 60, 70) millimeters. This method ensures that the coordinate transformation result is close to the target point through multiple iterations of optimization. For example, when updating the robot motion control parameters, control instructions can be generated based on the corrected target coordinates. Assuming that the robot uses joint angle control, the corrected coordinates (50, 60, 70) millimeters can be converted into joint angle adjustment amounts, such as joint 1 rotating 2 degrees and joint 2 rotating 1.5 degrees. After the robot is driven by the control instructions, real-time position feedback is obtained through the laser sensor, and the feedback coordinates are assumed to be (50.1, 60.2, 69.9) millimeters, and the position error is calculated as (0.1, 0.2, -0.1) millimeters. If the preset accuracy requirement is that the error is less than 0.5 millimeters, the requirement is met, indicating that the control instructions are effective. This process verifies the accuracy of the compensation vector through real-time feedback. In one possible implementation, if the position error does not meet the accuracy requirement, the compensation function can be further optimized.For example, by introducing a weighting factor to adjust the compensation vector, the correction weight of the key axis is increased, such as assigning a weight of 1.2 to the Z-axis error, generating a new compensation vector (2, 2, 2.4) mm. This method dynamically adjusts the weight to adapt to the error characteristics in different directions, improving the stability of the robot motion. Specifically, the design of the feedback mechanism can combine multi-sensor fusion, such as laser sensors providing high-precision distance data and vision sensors providing angle information, and generating comprehensive position feedback after fusion. This method complements multiple sources of data to improve the robustness of error judgment, thereby optimizing the generation efficiency of control instructions.

[0023] Step 5), verify the corrected coordinate conversion result using an iterative optimization algorithm, obtain the deviation data between the actual arrival position and the theoretical target position by executing the standard test action by the robot, and if the deviation exceeds the allowed range, return to step three to adjust the compensation model parameters again.

[0024] By executing the standard test action by the robot, actual arrival position data and theoretical target position data are obtained, deviation data between the two is calculated, and an iterative optimization algorithm is used to process the deviation data to generate a position deviation analysis result. Determine if the deviation exceeds the preset threshold. If the deviation exceeds the preset threshold, adjust the compensation model parameters using the gradient descent algorithm to generate updated compensation model parameters. Re-execute the coordinate conversion using the updated compensation model parameters to generate a new coordinate conversion result. Execute the standard test action again using the robot to obtain new actual arrival position data. Calculate the deviation data from the theoretical target position. If the new deviation data is still outside the preset threshold, repeat the gradient descent algorithm to adjust the compensation model parameters to obtain optimized compensation model parameters. Execute the final coordinate conversion using the optimized compensation model parameters to generate a verification result output.

[0025] In a possible implementation of the present application, the process of acquiring actual arrival position data and theoretical target position data by executing a standard test action by the robot can be understood as a basic link of precision verification. For example, assuming that an industrial robot needs to complete a point-to-point movement task in a two-dimensional plane, the standard test action can be designed to move from the origin (0, 0) to the target point (100, 100) millimeters. The actual arrival position is recorded by a high-precision laser range finder as (99.5, 100.2) millimeters, and the theoretical target position is (100, 100) millimeters. The deviation data is calculated by comparing the two coordinates, that is, the x-axis deviation is -0.5 millimeters, and the y-axis deviation is 0.2 millimeters. The acquisition of such deviation data provides a basis for subsequent analysis and emphasizes the importance of high-precision measurement equipment in data collection. For example, when processing deviation data using an iterative optimization algorithm, an optimization method based on the least squares method can be selected. Specifically, the system will aggregate the deviation data of multiple test actions, for example, the x-axis deviations of 10 tests are -0.5, -0.4, -0.6 millimeters, etc., forming a deviation data set. The iterative optimization algorithm analyzes these data to determine the distribution of the deviation and generates a position deviation analysis result. Assuming that the analysis result shows that the mean x-axis deviation exceeds the preset threshold of 0.3 millimeters, the subsequent compensation model adjustment is triggered. This method ensures the comprehensiveness of deviation analysis and helps to discover systematic errors. In an embodiment of the present application, if the deviation exceeds the preset threshold, the gradient descent algorithm is used to adjust the compensation model parameters. For example, the initial parameters of the compensation model may include the scale factor and the offset of coordinate conversion, which are 1.0 and 0 millimeters, respectively. Based on the deviation data, the gradient descent algorithm adjusts the scale factor to 1.005 and the offset to -0.2 millimeters through multiple iterations. This adjustment process reduces the deviation by gradually optimizing and ensures that the model is closer to the actual motion characteristics. The optimized parameters can improve the accuracy of coordinate conversion. Specifically, by using the updated compensation model parameters to perform coordinate conversion again, the robot calculates the target coordinates based on the new scale factor and offset. For example, the input target point (100, 100) millimeters is converted to (100.5, 99.8) millimeters by the adjusted model. This new coordinate conversion result is closer to the actual demand, demonstrating the effectiveness of model optimization. For example, when the robot performs the standard test action again to verify the new model, the actual arrival position may be recorded as (100.1, 100.0) millimeters, and the new deviation is x-axis 0.1 millimeter and y-axis 0.0 millimeter, both of which are within the preset threshold of 0.3 millimeters. This verification process confirms the reliability of model adjustment and reduces the need for repeated optimization. In a possible implementation of the present application, if the new deviation data still exceeds the threshold, the gradient descent algorithm is repeated. For example, when the x-axis deviation is 0.4 millimeters, the system further adjusts the scale factor to 1.006 and the offset to -0.25 millimeters.The persistence of this iterative optimization ensures the gradual convergence of the deviation, improves the robustness of the system, and finally, the final coordinate transformation is performed based on the optimized compensation model parameters to generate verification result output, for example, the robot moves to (100.05, 99.95) millimeters based on the final parameters, with a deviation of only 0.05 millimeters, meeting the high precision requirement. This method ensures the stability and accuracy of coordinate transformation through multiple rounds of verification and optimization, providing reliable support for robot motion control.

[0026] Step 6), establish a calibration accuracy evaluation system based on the verification results, calculate the repeatability and absolute positioning accuracy of the system by statistical analysis of the position deviation data of multiple tests, and obtain the system performance indicators under the current calibration state.

[0027] Collect position deviation data of multiple tests through sensors, use mean and standard deviation calculation methods to obtain data distribution characteristics, use normal distribution model to analyze position deviation data based on data distribution characteristics, determine the statistical characteristics of repeatability, if the statistical characteristics of repeatability exceed the preset threshold, optimize the calibration parameters through the least squares method to obtain the adjusted calibration model, execute the test data collection again through the adjusted calibration model to obtain new position deviation data, use the variance analysis method to calculate the change trend of absolute positioning accuracy based on the new position deviation data, if the change trend of absolute positioning accuracy does not reach the preset standard, further optimize the calibration parameters through the gradient descent algorithm to obtain the optimized calibration model, execute the final test data collection through the optimized calibration model to obtain the system performance indicators.

[0028] In a possible implementation manner of the present application, the position deviation data collected by the sensor in multiple tests is a basic link to ensure the accuracy of robot motion control. For example, when an industrial robot performs a point-to-point movement task, the sensor can be a high-precision encoder installed at the joint of the robot to record the actual position of each movement. Assuming that the robot moves 100 millimeters from the origin to the target point, the sensor collects 10 test data continuously, and records the actual positions as 99.8 millimeters, 100.1 millimeters, 99.7 millimeters, etc. This data collection method emphasizes the stability and high-frequency sampling capability of the sensor, which helps to capture the subtle changes of the position deviation and provides a reliable data basis for subsequent analysis. For example, when the mean and standard deviation calculation method is used to analyze the data distribution characteristics, the system will summarize the deviation data of multiple tests. Assuming that the deviation data of 10 tests is-0.2 millimeters, 0.1 millimeters, -0.3 millimeters, etc., the mean is-0.1 millimeters and the standard deviation is 0.15 millimeters. These statistical indicators reflect the concentration trend and dispersion degree of the deviation. The mean indicates that there is a slight negative deviation in the system, and the standard deviation shows the fluctuation range of the data. This method quantifies the deviation distribution to provide an intuitive basis for subsequent modeling. In a possible implementation manner, when the position deviation data is analyzed based on the normal distribution model, the system assumes that the deviation data conforms to the normal distribution characteristics. For example, based on the mean-0.1 millimeters and the standard deviation 0.15 millimeters, the system can draw the probability density curve of the deviation to determine whether the deviation range in the 95% confidence interval meets the preset threshold of 0.2 millimeters. If it is found that the deviation range exceeds the threshold, it means that the repeatability is insufficient. This analysis method reveals the regularity of the deviation through the statistical model to provide theoretical support for the optimization of calibration parameters. For example, when the least squares method is used to optimize the calibration parameters, the system adjusts the model parameters based on the deviation data set. Assuming that the calibration model contains a scale factor with an initial value of 1.0, the system adjusts the scale factor to 1.003 through the least squares method based on the deviation data. This method minimizes the sum of squares of deviations to ensure that the calibration model is closer to the actual motion characteristics. The optimized calibration model can reduce systematic errors and improve positioning stability. In a possible implementation manner, when the test data collection is re-executed, the robot uses the adjusted calibration model. For example, the target point is still 100 millimeters, and the new actual positions recorded by the sensor are 99.9 millimeters, 100.0 millimeters, etc. The deviation data is significantly reduced. This re-collection process verifies the effectiveness of the calibration model and provides data support for further analysis. For example, when the variance analysis method is used to calculate the variation trend of the absolute positioning accuracy, the system compares the deviation variances of multiple tests.Assuming that the variance of the new bias data decreases from 0.0225 to 0.01, indicating that the positioning accuracy tends to be stable, this analysis method quantifies the dispersion of the bias and clearly shows the effect of the calibration model optimization. In one possible implementation, if the absolute positioning accuracy is not up to standard, the system further optimizes the calibration parameters through the gradient descent algorithm. For example, based on the new bias data, the scale factor is fine-tuned from 1.003 to 1.004. This iterative optimization method gradually adjusts the parameters and gradually converges the bias to ensure the robustness of the model. For example, the final test data collection is based on the optimized calibration model, and the bias data recorded by the sensor may be 0.05 mm, -0.03 mm, etc. The system performance indicators show that the bias is within 0.1 mm. This result verifies the high precision of the optimized model and provides a reliable guarantee for robot motion control.

[0029] Step 7), an adaptive update mechanism is established for system performance indicators. When environmental temperature changes or mechanical structure deformation is detected, local update of calibration parameters is automatically triggered to adjust the compensation model through incremental learning to avoid full recalibration.

[0030] The environmental temperature data and mechanical structure deformation data are obtained in real time by the sensor, the temperature change rate and deformation degree are calculated, and the environmental change characteristics are obtained. If the environmental change characteristics exceed the preset threshold, the adaptive calibration process is triggered to determine the set of calibration parameters that need to be updated. The incremental learning algorithm is used to update the compensation model, the feature vector is extracted from the set of calibration parameters, and the updated model parameters are obtained. The compensation model is adjusted by the updated model parameters to generate new control signals, which are output to the system execution unit. Feedback data is obtained from the system execution unit, the performance indicator bias is calculated, and it is judged whether the stable control requirement is met. If the performance indicator bias exceeds the preset range, the calibration parameters are iteratively optimized to obtain new environmental change characteristics. According to the new environmental change characteristics, the above process is repeated to output the final system control signal.

[0031] For example, in a precision machining scenario, real-time acquisition of environmental temperature data and mechanical structure deformation data through sensors can effectively capture dynamic changes during system operation. Environmental temperature data is typically collected by high-precision thermocouple sensors. Assuming that a certain machining equipment is working, the environmental temperature rises from 20°C to 25°C, and the temperature change rate is 0.5°C / min. At the same time, the mechanical structure deformation data is measured by a laser displacement sensor, and the spindle deformation is detected to be 0.02 mm. These data together constitute the environmental change characteristics, reflecting the potential impact of temperature on mechanical precision. It should be noted that the temperature change rate and the deformation degree are key indicators for evaluating system stability. If the temperature change rate exceeds 0.3°C / min or the deformation exceeds 0.015 mm, the system may cause machining errors due to thermal expansion or mechanical stress. In one possible implementation, if the environmental change characteristics exceed the preset threshold, for example, the temperature change rate reaches 0.4°C / min, the adaptive calibration process is triggered. The calibration parameter set may include the thermal compensation coefficient of the spindle and the control gain of the servo motor. Assuming that through analysis, it is determined that the thermal compensation coefficient needs to be adjusted from 0.8 to 0.85 to offset the expansion caused by temperature rise. When updating the compensation model using an incremental learning algorithm, the system extracts feature vectors from historical data, such as the correlation coefficient of temperature change rate and deformation, to generate new model parameters. The advantage of this approach is that it can dynamically adapt to environmental changes and avoid frequent recalibration. For example, after updating the model parameters, the new control signal is generated to adjust the spindle position by driving the servo motor. Assuming that the spindle offset is reduced from 0.02 mm to 0.01 mm after adjustment, it indicates that the control signal effectively reduces the deformation impact. After obtaining feedback data from the system execution unit, the performance index deviation is calculated, for example, the machining part size error is reduced from 0.03 mm to 0.01 mm, which meets the stable control requirements. If the deviation is still out of range, for example, the size error is 0.02 mm, the calibration parameters are iteratively optimized, and the environmental change characteristics are recalculated. Further preferably, iterative optimization can be based on the temperature-deformation mapping relationship in historical data to adjust the compensation coefficient to 0.87, further reducing the error. In one embodiment, the new environmental change characteristics may show that the temperature change rate is reduced to 0.2°C / min and the deformation is stable at 0.01 mm. At this time, the system repeats the above process to output the final control signal, ensuring long-term stability of machining precision. The advantage of this method is that through continuous monitoring and dynamic adjustment, the system can maintain high-precision operation in complex environments, significantly improving machining quality and equipment reliability.

[0032] Step 8), by monitoring the coordinate transformation accuracy change in the robot operation process in real time, using a sliding window algorithm to analyze the positioning error trend of the last several operations, if the error shows systematic growth, the automatic optimization process of the calibration parameters is started to ensure long-term stable operation of the system.

[0033] The coordinate conversion data in the robot operation process is collected in real time by a sensor, stored as a time series data set, and the original coordinate data set is obtained; the original coordinate data set is processed by a sliding window algorithm, the mean and standard deviation of the positioning error in a specified time window are calculated, and the error statistical characteristics are determined; if the mean in the error statistical characteristics exceeds the preset threshold and the standard deviation continues to increase, it is judged that the error presents a systematic growth, and a trigger signal is generated; according to the trigger signal, the initial value of the calibration parameter is obtained from the preset parameter library, the gradient descent algorithm is used to optimize the calibration parameter, and the optimized parameter set is obtained; the robot coordinate conversion model is updated by the optimized parameter set, the positioning error is recalculated, the updated error data is obtained, the sliding window algorithm is used to analyze the updated error data, and if the error mean is lower than the preset threshold, it is confirmed that the system returns to the stable operation state, the optimized operation state is continuously monitored by time series analysis, and if a new error growth trend is detected, the above optimization process is repeated to obtain a stable system operation state.

[0034] For example, in the process of robot operation, real-time acquisition of coordinate transformation data by sensors is the basis of the entire process. Sensors can be deployed on key joints or moving parts of robots to continuously record changes in position coordinates, forming a time series data set. For example, assume that an industrial robot is used for precise positioning tasks on an assembly line, and the sensor collects 10 coordinate data per second, recording the three-dimensional coordinate values of the robot end effector. These data are stored as a time series data set, providing the original basis for subsequent analysis. It can be understood that the integrity and accuracy of the time series data are crucial for subsequent error analysis. In one embodiment, when processing the original coordinate data set using a sliding window algorithm, a time window of 5 seconds covering 50 data points can be set. The sliding window algorithm calculates the mean and standard deviation of the positioning error in each window by moving the window sequentially. For example, the mean error recorded in a certain window is 0.5 mm, and the standard deviation is 0.2 mm. It should be noted that the mean reflects the overall deviation level of the error, while the standard deviation indicates the dispersion of the error. If the mean gradually increases and the standard deviation continues to expand, it may mean that the robot positioning system is disturbed by external factors or internal parameter drift. For example, if the mean error in the error statistics exceeds the preset threshold of 0.4 mm, and the standard deviation increases from 0.2 mm to 0.3 mm and continues to rise, it is judged that the error is showing systematic growth. Systematic growth may be caused by environmental vibration, load change, or mechanical wear and tear. In one possible implementation of the present application, the system generates a trigger signal to prompt the need to adjust the coordinate transformation model. It can be understood that timely detection of systematic error growth trends helps to avoid further deterioration of positioning accuracy. In one embodiment, when obtaining the initial value of the calibration parameter from the preset parameter library, the appropriate parameter set can be selected according to the current task type of the robot. For example, for high-precision assembly tasks, the initial values stored in the parameter library may be empirical values based on historical calibration data, such as a joint angle offset of 0.1 degrees and a coordinate scaling factor of 1.02. When using the gradient descent algorithm to optimize these parameters, the system gradually adjusts the parameter values to gradually reduce the positioning error. It can be understood that the gradient descent algorithm finds the optimal solution for error through iteration, which helps to quickly converge to the appropriate parameter set.For example, after updating the robot coordinate transformation model with the optimized parameter set, the positioning error is recalculated. Assuming that the average error in a certain time window after updating is reduced from 0.5 mm to 0.3 mm, which is lower than the preset threshold of 0.4 mm, it should be noted that the reduction of the average error indicates that the coordinate transformation model has better adapted to the current working state. In one possible implementation, the system continuously analyzes the updated error data through the sliding window algorithm, confirms that the average error is stable at about 0.3 mm, and the standard deviation no longer significantly increases, indicating that the system has recovered stable operation; in an embodiment of the present application, time series analysis is used to continuously monitor the optimized running state; for example, the system generates an error trend report every minute to observe the changes of the average error and the standard deviation. If a new error growth trend is detected, such as the average error rising from 0.3 mm to 0.45 mm and the standard deviation increasing from 0.15 mm to 0.25 mm, it indicates that there may be new interference factors. It should be noted that continuous monitoring can timely discover potential problems and trigger a new round of optimization process. For example, the system again obtains initial values from the parameter library and optimizes them in combination with the current error data to adjust the coordinate transformation model until the average error is lower than the threshold again. It can be understood that the above process forms a closed-loop mechanism through real-time data collection, error analysis, parameter optimization and continuous monitoring. For example, the data collected by the sensor provides the basis for error analysis, the sliding window algorithm extracts error features, triggers a signal to start the optimization process, the gradient descent algorithm adjusts the parameters, and the model is updated and verified for stability through time series analysis. In one possible implementation, if the robot runs in a high-temperature or high-load environment for a long time, the system can dynamically adjust the threshold according to historical data, such as adjusting the average error threshold from 0.4 mm to 0.5 mm to adapt to environmental changes. It should be noted that this adaptive mechanism can flexibly cope with positioning requirements in different working conditions.

[0035] Obviously, those skilled in the art can make various modifications and variations to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A hand-eye calibration method suitable for a 5-axis hybrid serial-parallel manipulator, characterized in that, The method comprises: Step 1), acquiring coordinate data of a robot end effector at different positions and attitudes by using a multi-point sampling calibration algorithm, synchronously collecting corresponding image feature point coordinates by using a camera, and establishing an initial hand-eye calibration matrix, if the number of collected points is less than a preset threshold, then continue to increase the sampling points until the matrix calculation accuracy requirement is met; Step 2), analyzing the coordinate conversion error distribution characteristics according to the initial hand-eye calibration matrix calculation result, separating the total error into translation error and rotation error two components by using an error vector decomposition algorithm, and obtaining the distribution law and variation trend of the error in different spatial directions; Step 3), establishing a dynamic compensation model according to the error variation trend, fitting the functional relationship between the error and the robot operation distance and angle by using a least square method, if the fitting residual error exceeds the preset accuracy range, then adjusting the fitting order to recalculate until the compensation function meeting the accuracy requirement is obtained; Step 4), correcting the original hand-eye calibration matrix in real time by using the correction parameters calculated by the compensation function, determining the corresponding error compensation amount according to the current robot position and target point distance, and superimposing the compensation amount into the coordinate conversion process to obtain the corrected target coordinates; Step 5), verifying the corrected coordinate conversion result by using an iterative optimization algorithm, acquiring the deviation data of the actual reaching position and the theoretical target position by using the robot to perform a standard test action, if the deviation exceeds the allowable range, then returning to step three to adjust the compensation model parameters; Step 6), establishing a calibration accuracy evaluation system according to the verification result, calculating the repeat positioning accuracy and absolute positioning accuracy of the system by statistically analyzing the position deviation data of multiple tests, and obtaining the system performance index under the current calibration state; Step 7), establishing an adaptive updating mechanism for the system performance index, automatically triggering the local update of the calibration parameters when detecting that the environmental temperature changes or the mechanical structure slightly deforms, and adjusting the compensation model by using an incremental learning method to avoid overall recalibration; Step 8), monitoring the coordinate conversion accuracy change in the robot operation process in real time, analyzing the positioning error trend of the recent several operations by using a sliding window algorithm, if it is found that the error presents a systematic growth, then starting the automatic optimization process of the calibration parameters to ensure the long-term stable operation of the system.

2. The hand-eye calibration method suitable for a 5-axis serial-parallel hybrid robot arm according to claim 1, wherein, The step 1) further comprises: acquiring coordinate data of a robot end effector at different positions and attitudes by using a multi-point sampling calibration algorithm, synchronously collecting corresponding image feature point coordinates by using a camera, and obtaining an initial data set; processing the initial data set by using a least square method, constructing a hand-eye calibration matrix, and determining initial matrix parameters; if the calculation error of the initial matrix exceeds a preset threshold, then adjusting the sampling point position by using an iterative optimization algorithm, acquiring new coordinate data and feature point coordinates; updating the hand-eye calibration matrix according to the newly acquired coordinate data and feature point coordinates, and obtaining an optimized matrix; decomposing the optimized matrix by using a singular value decomposition algorithm, extracting main feature vectors, and judging whether the matrix accuracy meets the requirement; if the matrix accuracy is lower than the preset threshold, then adjusting the sampling point distribution strategy according to the distribution characteristics of the feature vectors, and acquiring supplementary coordinate data; Reconstruct the hand-eye calibration matrix by supplementing coordinate data to obtain the final calibration result.

3. The hand-eye calibration method for a 5-axis serial-parallel hybrid robot according to claim 1, wherein The step 2) further comprises: Obtain coordinate conversion error data through the initial hand-eye calibration matrix, separate the translation error and the rotation error by using the error vector decomposition algorithm to obtain error components; Extract the distribution characteristics of the translation error and the rotation error in the spatial coordinate system from the error components to determine the error distribution characteristics; If the error distribution characteristics exceed the preset threshold, perform spatial direction decomposition on the translation error and the rotation error to obtain the spatial direction distribution; Calculate the error vector modulus in each direction according to the spatial direction distribution to obtain the distribution rule extraction result; Perform dimensionality reduction processing on the distribution rule extraction result by using the principal component analysis algorithm to obtain the change trend analysis data; Fit the error change curve through the change trend analysis data to determine the change trend of the error in the spatial direction; Extract key feature points from the change trend to generate a feature description of the spatial direction error distribution.

4. The hand-eye calibration method suitable for a 5-axis serial-parallel hybrid robot arm according to claim 1, wherein, The step 3) further comprises: Obtain robot operation distance and angle data, collect real-time operation parameters by using sensors, remove noise through data preprocessing to obtain a standardized operation data set; According to the standardized operation data set, use the least squares method to construct an initial fitting function of operation distance and angle and error to determine the initial fitting parameters; For the initial fitting function, calculate the fitting residual value, and compare the residual with the preset accuracy threshold to determine whether the accuracy requirement is met; If the residual exceeds the preset accuracy threshold, adjust the fitting order, and calculate the fitting function again by using the least squares method to obtain updated fitting parameters; According to the updated fitting parameters, construct an error dynamic compensation model to generate a compensation function; Correct the real-time operation distance and angle data by using the compensation function to obtain corrected operation parameters; Update the robot control instructions by using the corrected operation parameters to obtain an optimized motion trajectory.

5. The hand-eye calibration method for a 5-axis serial-parallel hybrid robot according to claim 1 or 2 or 3, characterized in that, The step 4) comprises: Calculate the correction parameters through the error compensation function, and obtain the position error from the distance between the current position of the robot and the target point; If the position error exceeds the preset threshold, generate correction parameters according to the error compensation function; Adjust the hand-eye calibration matrix in real time by using the correction parameters to obtain an adjusted calibration matrix; Perform coordinate system transformation according to the adjusted calibration matrix to obtain initial conversion coordinates; Calculate the error compensation amount from the initial conversion coordinates and the distance to the target point; If the error compensation amount meets the dynamic adjustment condition, generate a compensation vector; Add the compensation vector to the initial conversion coordinates to obtain the modified target coordinates; Update the robot motion control parameters by using the modified target coordinates to generate control instructions; Drive the robot to move according to the control instructions to obtain real-time position feedback and determine whether the position error meets the accuracy requirement.

6. The hand-eye calibration method for a 5-axis serial-parallel hybrid robot according to claim 5, wherein The step 5) comprises: Perform a standard test action by the robot to obtain actual arrival position data and theoretical target position data, and calculate the deviation data between the two; Process the deviation data by using an iterative optimization algorithm to generate a position deviation analysis result and determine whether the deviation exceeds the preset threshold; If the deviation exceeds the preset threshold, a gradient descent algorithm is used to adjust the compensation model parameters to generate updated compensation model parameters; The coordinate conversion is re-executed through the updated compensation model parameters to generate a new coordinate conversion result; The robot is used to perform the standard test action again to obtain new actual arrival position data, and deviation data from the theoretical target position is calculated; If the new deviation data is still outside the preset threshold, the gradient descent algorithm is repeatedly executed to adjust the compensation model parameters to obtain optimized compensation model parameters; The final coordinate conversion is performed through the optimized compensation model parameters to generate a verification result output.

7. The hand-eye calibration method for a 5-axis serial-parallel hybrid robot according to claim 2, wherein The step 6) further includes: The position deviation data of multiple tests is collected by the sensor, and the data distribution characteristics are obtained by using the mean and standard deviation calculation method; According to the data distribution characteristics, the position deviation data is analyzed by using a normal distribution model to determine the statistical characteristics of the repeated positioning accuracy; If the statistical characteristics of the repeated positioning accuracy exceed the preset threshold, the calibration parameters are optimized by the least squares method to obtain adjusted calibration model parameters; The test data collection is re-executed through the adjusted calibration model to obtain new position deviation data; According to the new position deviation data, the change trend of the absolute positioning accuracy is calculated by using the variance analysis method; If the change trend of the absolute positioning accuracy does not reach the preset standard, the calibration parameters are further optimized by the gradient descent algorithm to obtain an optimized calibration model; The final test data collection is performed through the optimized calibration model to obtain system performance indicators.

8. The hand-eye calibration method suitable for a 5-axis serial-parallel hybrid robot arm according to claim 7, wherein, The step 7) further includes: The environmental temperature data and mechanical structure deformation data are obtained in real time by the sensor, the temperature change rate and deformation degree are calculated, and the environmental change characteristics are obtained; If the environmental change characteristics exceed the preset threshold, the adaptive calibration process is triggered to determine the set of calibration parameters that need to be updated; The compensation model is updated by using the incremental learning algorithm, the feature vector is extracted from the set of calibration parameters, and the updated model parameters are obtained; The compensation model is adjusted by using the updated model parameters to generate new control signals, which are output to the system execution unit; The feedback data is obtained from the system execution unit, the performance indicator deviation is calculated, and it is judged whether the stable control requirement is met; If the performance indicator deviation exceeds the preset range, the calibration parameters are iteratively optimized to obtain new environmental change characteristics; According to the new environmental change characteristics, the above process is repeated to output the final system control signal.

9. The hand-eye calibration method for a 5-axis serial-parallel hybrid robot according to claim 5, wherein, The step 8) further includes: The coordinate conversion data in the robot operation process is collected in real time by the sensor and stored as a time series data set to obtain an original coordinate data set; The original coordinate data set is processed by using a sliding window algorithm to calculate the positioning error mean and standard deviation within a specified time window, and the error statistical characteristics are determined; If the mean in the error statistical characteristics exceeds the preset threshold and the standard deviation continuously increases, it is judged that the error presents systematic growth, and a trigger signal is generated; According to the trigger signal, the initial values of the calibration parameters are obtained from the preset parameter library, the calibration parameters are optimized by using the gradient descent algorithm, and an optimized parameter set is obtained; The robot coordinate conversion model is updated by using the optimized parameter set, the positioning error is recalculated, and updated error data is obtained; The updated error data is analyzed by using a sliding window algorithm, and if the error mean is lower than a preset threshold, it is confirmed that the system returns to a stable operation state; The optimized operation state is continuously monitored by time series analysis, and if a new error growth trend is detected, the above optimization process is repeated to obtain a stable system operation state.