Precise positioning guarantee system for intelligent analysis of mechanical hand grabbing

Abstract

The present application relates to the technical field of manipulator positioning control, and discloses a precision positioning guarantee system for intelligent analysis of manipulator grabbing, which comprises four modules of grabbing target recognition, grabbing stability evaluation, dynamic interference suppression and positioning compensation. The grabbing target recognition module collects surface profile data of the target to be grabbed, divides local areas and carries out cluster analysis to obtain candidate grabbing areas. The grabbing stability evaluation module calculates the surface fitting index according to the geometric feature variation trend of the candidate areas, and obtains the pressure attenuation and distribution difference in combination with the pressure related trend. The dynamic interference suppression module generates a dynamic interference coefficient in combination with multiple parameters, constructs a covariance matrix, and calculates the principal component contribution rate and the characteristic dispersion index. The positioning compensation module generates precision compensation parameters based on the correlation between rows of the covariance matrix, and corrects the trajectory of the manipulator end effector, thereby effectively improving the precision and stability of the manipulator grabbing positioning and adapting to complex grabbing environments.

Description

Technical Field

[0001] This invention relates to the field of robotic arm positioning and control technology, specifically to a precise positioning and assurance system for intelligent analysis of robotic arm grasping. Background Technology

[0002] In fields such as industrial automation, logistics sorting, and precision manufacturing, the accuracy of robotic arm grasping operations directly affects production efficiency and product quality. Currently, most robotic arm grasping systems rely on preset programs or simple visual recognition to achieve target positioning. When faced with grasping targets with complex surface shapes and subtle uneven features, it is difficult to accurately delineate the effective grasping area, often leading to slippage or target damage during the grasping process due to improper area selection.

[0003] Existing technologies for evaluating the gripping area often focus on single geometric parameters, such as area and curvature, neglecting the impact of geometric feature variations and height differences between adjacent areas on gripping stability. This fails to comprehensively reflect the gripping area's adhesion performance, leading to biased gripping stability assessments. Furthermore, in pressure distribution analysis, traditional methods only consider the magnitude of extreme pressure values, neglecting the gradient variation trend of pressure distribution at the center point and the distance variation trend between extreme pressure points. This makes it difficult to accurately identify differences in pressure attenuation and distribution, further reducing the reliability of gripping stability assessments.

[0004] In practical applications, the grasping environment often presents dynamic interference factors such as vibration, airflow disturbances, and slight target displacement. Existing systems lack effective methods for quantifying dynamic interference and cannot adjust the grasping strategy in real time according to the interference situation, resulting in a significant decrease in positioning accuracy due to environmental influences. Furthermore, in the positioning compensation stage, existing technologies mostly compensate based on a single error source, failing to comprehensively consider multi-dimensional factors such as the correlation of surface contour data, the degree of dynamic interference, and the contribution of principal component features. The generation of compensation parameters lacks comprehensiveness, making it difficult to achieve high-precision trajectory correction and failing to meet the stringent requirements of robotic arm grasping and positioning in fields such as precision manufacturing. These problems result in significant deficiencies in the adaptability, stability, and accuracy of existing robotic arm grasping systems, hindering their application expansion in complex scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a precise positioning and assurance system for intelligent analysis of robotic arm grasping, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a precise positioning and assurance system for intelligent analysis of robotic arm grasping, the system comprising: The target recognition module is used to collect surface contour data of the target to be grasped, divide local regions according to the distribution of concave and convex features in the surface contour data, and perform cluster analysis on the surface contour data based on the geometric feature differences of the local regions to obtain each candidate grasping region. The grasping stability assessment module is used to calculate the surface adhesion index of each candidate grasping region based on the geometric feature change trend of the candidate grasping region and the height difference fluctuation between adjacent regions; and to obtain the pressure attenuation difference and pressure distribution difference based on the pressure distribution gradient change trend of the center point of the candidate grasping region and the distance change trend between pressure extreme points. The dynamic interference suppression module is used to generate dynamic interference coefficients for the grasping environment by combining pressure attenuation differences, pressure distribution differences, the average surface adhesion index of all candidate grasping areas, and the average morphological similarity between candidate grasping areas; it constructs the covariance matrix of surface contour data, obtains the eigenvalue sequence through matrix decomposition, and calculates the principal component contribution rate and feature dispersion index based on the concentration and dispersion of the eigenvalues. The positioning compensation module is used to generate precision compensation parameters for grasping and positioning based on the inter-row correlation of the covariance matrix, dynamic interference coefficient, principal component contribution rate, and feature dispersion index; and to perform trajectory correction on the end effector of the robot based on the precision compensation parameters.

[0007] Preferably, the method for obtaining each candidate crawling region is as follows: An edge detection algorithm is used to extract all convex and concave points in the surface contour data. The surface contour data is divided into several sub-regions with each concave point as the dividing boundary. The variance of the geometric features in each sub-region is calculated, and the variance sequence is input into a clustering algorithm to obtain a classification cluster set. The cluster with the largest variance of geometric features within the classification cluster is selected as the preferred cluster, and the sub-regions corresponding to the preferred clusters are marked as candidate crawling regions.

[0008] Preferably, the method for calculating the surface adhesion index of each candidate grasping region is as follows: The region symmetry coefficient and the contact surface adaptation coefficient are calculated based on the fluctuation amplitude of local geometric features in the candidate grasping region and the cumulative change of the height difference between adjacent regions. The product of the regional symmetry coefficient and the contact surface adaptation coefficient is used as the surface adhesion index of the candidate gripping region.

[0009] Preferably, the method for calculating the regional symmetry coefficient and the contact surface adaptation coefficient is as follows: For each candidate capture area, the highest point of the area and its continuous contour points to the left of the area are extracted to form the left contour segment, and the continuous contour points to the right of the highest point are formed the right contour segment. Calculate the first-order difference sequences of the left and right contour segments respectively, and generate a symbolic difference sequence by processing with a symbolic function; The ratio of the absolute value of the cumulative symbolic difference sequence of the left contour segment to the length of the left contour segment is used as the left fitness. The right fitness is calculated using the same method. The harmonic mean of the left fitness and the right fitness is taken as the contact surface fitness coefficient.

[0010] Preferably, the method for obtaining the pressure distribution difference is as follows: The pressure values ​​of the center points of all candidate grabbing areas are sorted by spatial location to generate a pressure sequence, and the coordinates of each pressure extreme point are sorted by distance to generate a location sequence. Calculate the first-order difference sequences of the pressure sequence and the position sequence respectively. Use the variance of the pressure sequence difference sequence as the pressure attenuation difference and the variance of the position sequence difference sequence as the pressure distribution difference.

[0011] Preferably, the method for generating the dynamic interference coefficient is as follows: The arithmetic mean of the surface adhesion index of all candidate gripping regions is calculated as the global adhesion benchmark. The mean Euclidean distance similarity between each candidate capture region and other regions is used as the region morphological similarity. The weighted sum of pressure attenuation difference, pressure distribution difference, global fit benchmark, and mean regional morphological similarity is used as the dynamic interference coefficient.

[0012] Preferably, the method for calculating the principal component contribution rate is as follows: After sorting the feature values ​​in descending order, input them into an adaptive threshold segmentation algorithm to divide the main feature values ​​and secondary feature values. The proportion of the number of principal eigenvalues ​​to the total number of eigenvalues ​​is used as the principal component contribution rate. The product of the mean and variance of the principal eigenvalues ​​is used as the principal eigenvalue volatility, and the product of the mean and variance of the secondary eigenvalues ​​is used as the secondary eigenvalue volatility. The absolute value of the difference between the two is used as the eigenvalue dispersion index.

[0013] Preferably, the method for generating the precision compensation parameters is as follows: Calculate the mean absolute value of the cosine similarity between each row of data in the covariance matrix and other rows of data, and use it as the inter-row correlation. The square root of the product of principal component contribution rate, feature dispersion index, inter-row correlation and dynamic interference coefficient is used as the accuracy compensation parameter.

[0014] Preferably, the method for trajectory correction of the robot end effector based on accuracy compensation parameters is as follows: The compensation amount for the joint angle of the robot is generated based on the accuracy compensation parameters. The compensation amount and the accuracy compensation parameters are linearly related in the formula. The compensation amount is converted into the pose adjustment amount of the end effector through inverse kinematics calculation and then superimposed on the original motion trajectory.

[0015] Preferably, the system further includes: The real-time feedback module is used to collect contact force data and position deviation data during the actual grasping process after trajectory correction, update the local geometric feature distribution of the surface contour data, and input the updated data into the grasping target recognition module for iterative calculation to form a closed-loop positioning guarantee mechanism.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This robotic arm's intelligent analysis and precise positioning system collects surface contour data of the target to be grasped through a target recognition module. It divides local areas based on the distribution of concave and convex features in the surface contour data, and performs cluster analysis based on the geometric feature differences of the local areas to obtain candidate grasping regions. Compared with traditional methods that rely solely on preset programs or simple visual recognition, it can more accurately identify effective areas that meet grasping requirements. Especially for targets with complex surface morphology, it can effectively avoid grasping slippage or target damage caused by improper area selection, and improve the system's adaptability to grasping targets of different shapes.

[0017] In the gripping stability assessment stage, the system not only considers the geometric feature change trend of the candidate gripping area and the height difference fluctuation between adjacent areas to calculate the surface adhesion index, but also combines the pressure distribution gradient change trend at the center point and the distance change trend between pressure extreme points to obtain pressure attenuation difference and pressure distribution difference. Compared with the traditional assessment method that only focuses on a single geometric parameter or pressure extreme value, it realizes a multi-dimensional and comprehensive assessment of gripping stability, which can more realistically reflect the adhesion performance and pressure transmission characteristics of the candidate gripping area, provide a more comprehensive judgment basis for the optimal selection of the gripping area, reduce the stability judgment bias caused by a single assessment dimension, and further ensure the reliability of the gripping process.

[0018] The dynamic interference suppression module generates dynamic interference coefficients by comprehensively considering pressure attenuation differences, pressure distribution differences, the average surface adhesion index of all candidate grasping areas, and the average morphological similarity between candidate grasping areas. Simultaneously, it constructs the covariance matrix of surface contour data and obtains eigenvalue sequences through matrix decomposition, calculating the principal component contribution rate and feature dispersion index. This enables quantitative analysis of dynamic interference factors in the grasping environment, overcoming the limitations of traditional systems that lack effective dynamic interference quantification methods. It accurately captures the impact of environmental factors such as vibration and airflow interference on the grasping process, providing crucial interference parameter support for subsequent positioning compensation. This helps the system adjust its strategy in real time according to interference conditions, reducing the adverse effects of environmental factors on positioning accuracy.

[0019] The positioning compensation module generates accuracy compensation parameters based on the inter-row correlation of the covariance matrix, dynamic interference coefficient, principal component contribution rate, and feature dispersion index. Based on these parameters, it corrects the trajectory of the robot's end effector. Compared with traditional compensation methods that rely on a single source of error, this module fully integrates multi-dimensional information such as surface contour data features and dynamic interference levels. This makes the generation of compensation parameters more comprehensive and reasonable, enabling high-precision correction of the robot's end effector trajectory. This effectively improves the accuracy of grasping and positioning, meets the high-precision requirements of grasping operations in fields such as precision manufacturing, and expands the application scope of robot grasping systems in complex and high-precision scenarios. Attached Figure Description

[0020] Figure 1 This is a timing diagram of the precise positioning guarantee system for intelligent analysis of robotic arm grasping described in this invention. Figure 2 To obtain the working principle diagram of each candidate grabbing region; Figure 3 A schematic diagram illustrating the working principle for calculating the surface adhesion index of each candidate gripping region. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1This invention provides a precise positioning and assurance system for intelligent analysis of robotic arm grasping. The system includes: during system operation, a target recognition module collects surface contour data of the target to be grasped. This data is typically acquired by a high-precision 3D scanning device or a depth vision sensor and includes the object's 3D coordinate information and elevation features; based on the distribution of concave and convex features in the surface contour data, the system divides the overall contour into multiple local regions and extracts sub-regions with significant geometric feature differences as candidate grasping regions using a clustering analysis algorithm; a grasping stability evaluation module calculates the surface adhesion index of each region based on the geometric feature change trend of the candidate grasping regions and the height difference fluctuation between adjacent regions, while simultaneously analyzing the pressure distribution gradient change trend and pressure at the region's center point. The distance variation trend between extreme points is used to deduce the differences in pressure attenuation and pressure distribution. The dynamic interference suppression module integrates the differences in pressure attenuation, pressure distribution, the average surface adhesion index of all candidate grasping areas, and the average morphological similarity between areas to generate a dynamic interference coefficient reflecting the intensity of environmental disturbance. At the same time, it constructs the covariance matrix of the surface contour data, obtains the eigenvalue sequence through matrix eigenvalue decomposition, and then calculates the principal component contribution rate and the feature dispersion index. The positioning compensation module uses the inter-row correlation of the covariance matrix, the dynamic interference coefficient, the principal component contribution rate, and the feature dispersion index to generate a precision compensation parameter for correcting the positioning deviation of the robot arm. Finally, the motion trajectory of the end effector is adjusted in real time through this parameter to ensure the accuracy and stability of the grasping operation.

[0023] Example 1: See Figure 2 In acquiring candidate grasping regions, the system first collects surface point cloud data of the object to be grasped using a high-precision optical scanner or structured light sensor. This data contains dense 3D coordinate information and normal vector data of the object's surface, forming a complete surface contour dataset. The edge detection algorithm uses an improved Canny operator, which eliminates noise interference while preserving contour details through adaptive Gaussian filtering. Feature points are identified by calculating the elevation gradient magnitude and direction in each neighborhood of the point cloud. Concave points are defined as points with abrupt gradient direction changes and elevation values ​​lower than the threshold of the surrounding neighborhood, while convex points are regions with continuous gradient directions and elevation values ​​significantly higher than the surrounding background. The algorithm traverses all data points and marks the spatial locations of these feature points. Using all identified concave points as natural segmentation boundaries, the system divides the surface contour data into multiple continuous sub-regions. Each sub-region consists of a continuous point cloud sequence between adjacent concave points. The segmentation process ensures that each sub-region does not contain any concave points and is only bounded by concave points. When calculating the variance of geometric features within each sub-region, the system extracts the elevation value sequence, local curvature sequence, and surface normal vector angle sequence for each sub-region. After standardizing these feature sequences, the system calculates their multidimensional variance, which comprehensively reflects the degree of fluctuation of the geometric characteristics within the sub-region.

[0024] After arranging the variance values ​​of all sub-regions in spatial order to form a variance sequence, the system uses the DBSCAN algorithm based on density clustering for processing. The algorithm identifies dense regions in the variance sequence by setting neighborhood radius and minimum sample number parameters, grouping sub-regions with similar and continuously distributed variance values ​​into the same cluster. From the generated cluster set, the cluster with the largest geometric feature variance within the cluster is selected as the preferred cluster, because the sub-regions corresponding to this cluster have the most significant internal feature changes. This characteristic helps the robotic arm generate greater contact friction and adaptability during grasping. Finally, these sub-regions are marked as candidate grasping areas and transmitted to subsequent processing units. During implementation, the point cloud data acquisition frequency is synchronized with the robotic arm's movement speed to ensure the real-time nature of the contour data. The parameters of the edge detection algorithm are dynamically adjusted according to the material characteristics of the object's surface; a higher gradient threshold is used for reflective surfaces to avoid false detections. The concave point recognition stage excludes false concave points caused by noise, verifying their authenticity by checking the elevation continuity of the surrounding area. Only concave points confirmed by multiple consecutive neighborhoods are used as segmentation boundaries.

[0025] During sub-region partitioning, the system records the start and end coordinates of each sub-region and establishes a region index table for rapid subsequent retrieval. Geometric feature variance is calculated point-by-point within each sub-region using a sliding window approach, with the window size adaptively varying according to the sub-region size. In the clustering analysis phase, the algorithm considers not only the similarity of variance values ​​but also the spatial relationships between sub-regions. Adjacent sub-regions with similar variances are preferentially grouped into the same cluster, ensuring the continuity of candidate regions. The selection of optimal clusters considers not only the maximum intra-cluster variance but also the cluster size and shape, prioritizing clusters with a moderate number of sub-regions and a concentrated spatial distribution, while avoiding overly dispersed clusters or those containing too few sub-regions. The final output candidate capture region includes its geometric feature descriptor, including information such as average elevation, curvature distribution range, and surface orientation.

[0026] The entire processing adopts a pipelined architecture, with each processing step executed in parallel to reduce latency. All algorithm modules are embedded and optimized to adapt to the resource constraints of the real-time operating system. Shared memory is used when data is transferred between different modules to avoid unnecessary copying overhead. The system also has error detection capabilities; when an anomaly occurs in the feature calculation of a sub-region, it automatically triggers recalculation or removes that region, ensuring the reliability of the output results. During implementation, the system loads different parameter configuration files for different types of crawling objects. For rigid objects, a stricter concave point detection threshold is used, while for flexible objects, the variance clustering conditions are relaxed. This adaptive capability allows the system to handle diverse crawling scenarios. The final generated candidate crawling region set is sorted by priority, which is comprehensively evaluated based on region size, location accessibility, and feature saliency, for subsequent modules to select from.

[0027] Example 2: See Figure 3 In calculating the surface adhesion index of each candidate grasping area, the system first performs geometric feature analysis on each candidate area, extracting local geometric feature fluctuation data within the area. This data is obtained by calculating the sum of squares of the deviations between the elevation value of each point within the area and the average elevation of the area. Simultaneously, the system calculates the cumulative change in elevation difference between this area and adjacent areas, obtained by integrating the elevation difference scores of consecutive points on the boundary line of the region. The calculation of the regional symmetry coefficient requires first determining the position of the highest point in each candidate grasping area. The highest point is defined as the point with the largest elevation value within the area, and the elevation changes in its surrounding neighborhood are radially distributed. A continuous sequence of points is extracted from the highest point to the left along the contour line to form the left contour segment, and a continuous sequence of points is extracted to the right to form the right contour segment. The lengths of the left and right contour segments are dynamically determined based on the total length of the area, while maintaining an equal number of points on both sides.

[0028] The first-order difference values ​​of the elevation sequence for the left and right contour segments are calculated separately. The first-order difference reflects the elevation change rate between adjacent points. A sign function is applied to the difference values ​​to convert positive values ​​to 1, negative values ​​to -1, and zero values ​​to 0, generating a symbolic difference sequence. The sum of the absolute values ​​of the symbolic difference sequence for the left contour segment is accumulated and divided by the number of points in the left contour segment to obtain the left fitness. Similarly, the corresponding value for the right contour segment is calculated as the right fitness. The harmonic mean of the left and right fitness is taken as the contact surface fitness coefficient. The harmonic mean is calculated using standard mathematical methods. The calculation of the region symmetry coefficient is based on the Fourier descriptors of the left and right contour segments. First, a Fourier transform is performed on the elevation sequence of the left and right contour segments to obtain the phase spectrum. The cosine similarity of the phase spectra on both sides is calculated as a symmetry measure. This similarity value is normalized and used as the region symmetry coefficient. Finally, the region symmetry coefficient is multiplied by the contact surface fitness coefficient to obtain the surface adhesion index, which is limited to the range of 0 to 1 for subsequent comparisons.

[0029] The system first performs coordinate standardization on the candidate capture regions, transforming the region point cloud into a local coordinate system with the highest point as the origin. This eliminates the impact of positional offset on the calculation. The calculation of local geometric feature fluctuation amplitude uses a sliding window approach, with the window size adaptively adjusted according to the region's radius of curvature. For regions with greater curvature, a smaller window is used to capture detailed changes. The calculation of the cumulative change in height difference between adjacent regions is performed along the region boundary line, which is discretized into a series of equally spaced points. The elevation difference between each discrete point and its corresponding point on the opposite side is calculated, and these differences are integrated to obtain the total change. The extraction of left and right contour segments follows the natural direction of the contour lines. When the contour lines have branches, the main branch is selected as the extraction target. During the extraction process, curve smoothing is performed to eliminate jagged fluctuations caused by noise.

[0030] First-order difference calculation employs the central difference method to improve accuracy. A small threshold is set during sign function processing to avoid misclassification due to numerical errors. Before absolute value accumulation, the difference sequences are standardized to ensure they are of the same order of magnitude. The harmonic mean calculation prioritizes the influence of smaller values, which helps identify regions with poor adaptability and prevents the good adaptability of a single aspect from masking overall defects. Before Fourier transform, the high-order sequence is interpolated to ensure the left and right contour segments have the same number of points. Phase spectrum comparison only uses the first few low-frequency components to improve computational efficiency and noise resistance. Similarity calculation uses a cosine similarity metric and undergoes a linear transformation to the 0-1 range. Finally, the surface fit index is calculated using floating-point multiplication, retaining sufficient decimal places to ensure accuracy. All calculations employ fixed-point arithmetic to accommodate the resource limitations of the embedded platform.

[0031] The system independently calculates the surface fit index for each candidate grasping region. The calculation process is fully parallelized to improve efficiency, and intermediate results are cached for possible subsequent reuse. When the region shape is exceptionally complex, the system activates an alternative calculation strategy, employing a spline curve-based fitting method instead of directly using the raw point cloud data, which improves computational stability. The calculation process includes multiple error detection mechanisms, including numerical range checks, convergence verification, and result consistency checks. When anomalies are detected, recalculation is automatically triggered or the region is marked as unusable. All algorithm parameters can be adjusted via configuration files, using different parameter combinations for grasping objects of different materials. For example, for soft objects, symmetry requirements are reduced while adaptability weights are increased. The final generated surface fit index is stored in the database along with other region feature indicators, sorted from highest to lowest index value. Regions with higher index values ​​are considered more suitable as final grasping points, and this data provides important input for subsequent grasping stability evaluation.

[0032] In the automated gripping scenario of automotive engine cylinder blocks, the system needs to calculate the surface fit index of the camshaft bore edge area on the top of the cylinder block. This area contains complex casting features and mounting holes. A 3D scanner acquires point cloud data of the upper surface of the cylinder block at a resolution of 0.1mm, obtaining a surface contour dataset containing 32,768 coordinate points. The area surrounding the camshaft bore is identified as a candidate gripping region, which exhibits a ring-shaped flange structure with local concave features. The system extracts the local geometric feature fluctuation amplitude data within this candidate region. By calculating the sum of squares of the elevation deviations of each point within the ring-shaped region from the reference plane, the fluctuation amplitude value is obtained as 0.87. Simultaneously, the cumulative change in the height difference between this region and the adjacent cylinder mounting surface is calculated. Integrating the height difference values ​​of 128 discrete points along the boundary line yields a change of 3.45mm.

[0033] The calculation of the regional symmetry coefficient first determines the highest point of the annular region, located at the flange vertex in the due north direction. From this point, 42 consecutive points are extracted along the contour line to the left to form the left contour segment, and the same number of points are extracted to the right to form the right contour segment. Both contour segments contain undulating casting features. After Gaussian filtering and smoothing, the high-order sequence of the left and right contour segments is used to calculate the first-order difference value sequence. The difference value of the left contour segment fluctuates within ±0.3 mm, and the difference value of the right contour segment fluctuates within ±0.4 mm. The sign function processing converts the difference value of the left contour segment into a sign sequence containing 28 ones and 14 ones, and the right contour segment generates a sign sequence containing 31 ones and 11 ones. The fitness of the left contour segment is calculated by dividing the sum of the absolute values ​​of the sign sequences (42) by the contour segment length (42), resulting in 1.0. The fitness of the right contour segment is calculated similarly and also results in 1.0. The harmonic mean is calculated as 1.0 and used as the contact surface fitness coefficient.

[0034] Fourier descriptor calculations performed a 256-point Fast Fourier Transform on the high-order sequences of the left and right contour segments. Cosine similarity was calculated using the first 16 low-frequency phase components. The phase spectrum of the left contour segment primarily concentrated in the 4th, 8th, and 12th harmonics, while the phase spectrum of the right contour segment distributed in the 4th, 7th, and 11th harmonics. The cosine similarity of the phase spectra of both sides was calculated to be 0.83, which was normalized and used as the region symmetry coefficient. The final surface fit index was calculated to be 0.83 × 1.0 = 0.83, and this value was marked as the fit quality index for this candidate region. The system adjusted the calculation parameters for the cast iron engine block. The Gaussian filter window size was set to 5 points to accommodate the roughness of the cast surface, and the threshold for the sign function processing was set to 0.05 mm to ignore minor vibration noise. Before the Fourier transform, the contour segment data underwent Hanning window processing to reduce spectral leakage. When comparing the phase spectra, the weighted focus was placed on the 4th to 12th harmonic components, which correspond to the periodic characteristics of the flange structure.

[0035] When processing points 23 to 28 of the right contour segment, abnormal fluctuations were detected. The system initiated a local resampling mechanism, increasing the sampling density around these points and recalculating the difference values ​​to ensure the accuracy of the symbol sequence. The entire calculation process took 8.7 ms, meeting the system's real-time requirements. The calculation results were then exchanged with the subsequent gripping stability evaluation module. To address potential oil stains or machining marks on the cylinder surface, the system implemented an adaptive adjustment mechanism. When abnormal elevation values ​​were detected in a local area, the filtering window was automatically expanded, and a hysteresis threshold was introduced in the symbol function processing to avoid frequent state switching. All intermediate calculation results were timestamped and quality-indicated for verification by subsequent modules. The final generated surface fit index was compared with the indices of other candidate regions. This annular region was selected as the priority gripping region due to its high symmetry and adaptability, and the relevant parameters were transmitted to the robot control system to guide the execution of the gripping action.

[0036] Example 3: The pressure values ​​at the center points of all candidate grasping areas are read. These pressure values ​​originate from a multi-point pressure sensor array installed at the end effector of the robotic arm. The sensors continuously collect pressure data at a sampling frequency of 500Hz. The determination of the center point relies on the calculation of the geometric center of the candidate area. This center is obtained by calculating the arithmetic mean of the point cloud coordinates of the area. The data from the sensor unit in the pressure sensor array closest to this geometric center is selected as the center point pressure value. All center point pressure values ​​are sorted according to their spatial location within their corresponding candidate areas. The spatial sorting follows the principle of continuity in the robotic arm's motion path, forming a pressure sequence. ,in Representing the Pressure measurements at the center point of each candidate region This represents the total number of candidate regions.

[0037] The identification of pressure extrema points employs a local maximum and minimum detection algorithm. This algorithm searches the pressure distribution map for points whose pressure values ​​are significantly higher or lower than their neighboring points. The coordinates of these extrema points are mapped to the three-dimensional workspace using the sensor unit number they belong to. These spatial coordinates of the extrema points are then sorted in ascending order of their Euclidean distance to form a location sequence. ,in Indicates the first The three-dimensional coordinate vector of each extreme point This represents the number of extreme points detected. Calculate the pressure sequence. First-order difference sequence This sequence characterizes the rate of change of pressure values ​​in space; similarly, the location sequence is calculated. First-order difference sequence ,in This represents the Euclidean norm, and the sequence characterizes the continuous variation in the distance between extreme points. Pressure sequence difference sequence. The variance was calculated as the pressure decay difference. ,in yes The mean value, which reflects the degree of fluctuation in pressure changes. Location sequence difference sequence. The variance was calculated as the pressure distribution difference. ,in yes The mean of the extreme points reflects the non-uniformity of the distribution of extreme points.

[0038] The generation of dynamic disturbance coefficients requires four inputs: pressure attenuation difference. Differences in pressure distribution Mean surface fit index of all candidate grabbing regions and the mean morphological similarity between candidate capture regions. Mean surface adhesion index The mean morphological similarity was calculated using a simple arithmetic mean. This is achieved by calculating the Euclidean distance between the feature vectors of each pair of candidate regions, converting it into a similarity score, and then averaging the scores of all pairs. The final dynamic interference coefficient is then calculated. From the formula:

[0039] in: , , , The preset weighting coefficients satisfy... The specific value of the weighting coefficient is dynamically adjusted based on the characteristics of the crawling task. For crawling tasks requiring high stability, the weighting coefficient will be increased. and The weight will be increased for tasks that need to adapt to changing environments. and The weight.

[0040] The data acquisition and processing of the pressure sensors are completely synchronized. A hardware triggering mechanism ensures the consistency of data timestamps. The determination of the center point includes outlier detection; when the data from a sensor deviates significantly from its neighborhood, a data repair procedure is initiated. The pressure sequence sorting algorithm considers the kinematic constraints of the robotic arm, ensuring that the sequence order is consistent with the actual path planning order. A minimum significance threshold is set for extreme point detection to avoid false extremes caused by noise. The first-order difference calculation uses the forward difference method, and the variance calculation uses an unbiased estimator. All calculations use floating-point operations to ensure accuracy, and intermediate results are cached for possible subsequent use. The calculation cycle of the dynamic interference coefficient is consistent with the system control cycle. The weighting coefficient is fine-tuned based on the output of the real-time environmental assessment module, which analyzes external factors such as vibration levels and changes in illumination. The processing flow includes multiple verification mechanisms, including numerical range checks, sequence length verification, and result rationality judgment. When abnormal pressure data or calculation errors are detected, valid data from the previous cycle is used or a recalculation is triggered. The finally generated dynamic interference coefficient is input into the positioning compensation module along with other system parameters to participate in the calculation of accuracy compensation parameters.

[0041] Example 4: The system first constructs a covariance matrix of the surface contour data. This matrix is ​​based on point cloud data of the object surface acquired by a high-precision 3D scanner. The point cloud data contains thousands of spatial coordinate points and their normal vector information. Taking a specific target to be grasped: the cylinder block of an automobile engine, its surface contains various geometric features, such as planes, grooves, and bolt holes. The system collects 1024 point cloud data points from the upper surface of the cylinder block as the input dataset. Each data point contains X, Y, and Z coordinate values ​​and NX, NY, and NZ normal vector components. When constructing the covariance matrix, the system combines the coordinates and normal vectors of each point into a 6-dimensional feature vector, calculates the covariance relationship between all pairwise feature vectors, and finally generates a 6×6 covariance matrix C. The rows and columns of this matrix correspond to the six dimensions of the feature vectors, and the matrix element C(i,j) represents the covariance value between the i-th feature and the j-th feature. Six eigenvalues ​​are obtained through matrix eigenvalue decomposition, and these eigenvalues ​​are arranged in descending order to form an eigenvalue sequence. .

[0042] Adaptive threshold segmentation employs a dynamic thresholding method based on gradient change. It calculates the gradient rate of change of the feature value sequence, and classifies primary and secondary features when the gradient value is less than a preset threshold. For engine block data, the feature value sequence is [4.32, 1.56, 0.87, 0.32, 0.18, 0.05], and the gradient sequence is [2.76, 0.69, 0.55, 0.14, 0.13]. When the gradient threshold is set to 0.5, the first three feature values ​​are determined as primary feature values.

[0043] Table 1: Eigenvalue distribution of the covariance matrix of surface profile data ; The principal component contribution rate is obtained by statistically analyzing the proportion of principal eigenvalues ​​to the total number of eigenvalues. In this example, the three principal eigenvalues ​​account for 50% of the total six eigenvalues, and the cumulative energy proportion of the principal eigenvalues ​​is calculated to be 91.7%. The calculation of the feature dispersion index requires calculating the mean of each principal eigenvalue. and variance The mean of the secondary eigenvalues and variance The principal characteristic fluctuation is calculated as follows: The secondary characteristic fluctuation is The absolute value of the difference between the two, 6.706, is used as the characteristic dispersion index.

[0044] The calculation of inter-row relevance requires analyzing the row vector similarity of the covariance matrix C and calculating the mean absolute value of the cosine similarity between each row and other rows. Taking row 1 as an example, its cosine similarities with rows 2 to 6 are 0.92, 0.85, 0.23, 0.17, and 0.09, respectively, with a mean absolute value of 0.45. The mean similarity of all rows is calculated using the same method, and the average is taken to obtain the inter-row relevance R = 0.38. The generation of accuracy compensation parameters requires integrating multiple calculation results, including the principal component contribution rate P = 0.50, the feature dispersion index E = 6.706, the inter-row relevance R = 0.38, and the dynamic interference coefficient D = 0.75 obtained from the dynamic interference suppression module. The system employs a dual verification mechanism to ensure calculation accuracy; all intermediate results undergo range verification and consistency checks. For different types of crawled objects, the system adaptively adjusts the gradient threshold and weight parameters; for example, it lowers the gradient threshold for smooth surface objects to capture more subtle feature changes. The calculation cycle is synchronized with the robot's control cycle to ensure that real-time performance requirements are met, while reserving sufficient computational margin to cope with sudden computational loads. The final generated accuracy compensation parameters, along with their corresponding confidence indices, are transmitted to the positioning compensation module for robot trajectory correction.

[0045] Example 5: The system operates using the task of grasping a car engine block as an example. This block has a complex geometry and multiple potential grasping planes. When the accuracy compensation parameter calculation module outputs a value of 0.977, this parameter is transmitted to the trajectory planning subsystem. The system converts the compensation amount into adjustment values ​​for the joint angles of the robotic arm based on a preset linear relationship model. The proportional coefficient in the linear relationship model is pre-calibrated to 0.45 based on the dynamic characteristics of the six-axis robotic arm, thereby generating the joint angle compensation amount. Radius. After receiving the joint angle compensation, the inverse kinematics calculation module uses a numerical iteration method to solve for the pose adjustment of the robot arm's end effector. This calculation process considers the current configuration state of the robot arm and the motion constraints. For the six-DOF robot arm in the example, the calculated position compensation amounts of the end effector in the X, Y, and Z directions are respectively... , , The attitude compensation is a rotation of 1.2 degrees around the Z-axis, and these adjustments are converted into homogeneous transformation matrix form.

[0046] The initial motion trajectory is pre-generated by the path planning module, containing a series of path points obtained through B-spline curve interpolation. Each path point records the target pose of the robotic arm's end effector in the task space. During the correction process, the system right-multiplies the transformation matrix corresponding to the pose adjustment amount with the pose matrix of each path point to achieve an overall trajectory offset. The corrected path point data is transmitted to the robotic arm controller via real-time Ethernet. The real-time feedback module is activated during the grasping action, collecting contact force data through a miniature force sensor array installed at the tip of the robotic finger at a sampling frequency of 1kHz. Simultaneously, a vision servo system detects the positional deviation between the end effector and the target object at a frequency of 100Hz. When an abnormal contact force distribution or a positional deviation exceeding a threshold is detected, the system triggers a data update mechanism, fusing the latest force sensor readings and visual measurement data into the surface contour database.

[0047] The surface contour data is updated using a weighted fusion strategy. Newly acquired data is assigned a weight coefficient of 0.7, while historical data has a weight of 0.3. The updated data focuses on correcting the distribution of local geometric features, particularly the curvature changes and normal vector directions in the contact area. The updated contour data is then re-input into the target recognition module, initiating a new round of candidate region segmentation and feature calculation, forming a closed-loop control process. The entire closed-loop system's operating cycle is synchronized with the robotic arm's control cycle, set at 10ms, completing the complete chain of data acquisition, processing, decision-making, and execution within each control cycle. The system maintains a dynamic priority queue to ensure that time-sensitive computational tasks are processed first, while a double-buffering mechanism avoids read / write conflicts during data updates.

[0048] To address the unique characteristics of engine block grasping tasks, the system incorporates anomaly handling routines. When oil or debris is detected on the block surface, the feedback gain is automatically increased to enhance the suppression of environmental interference. All computational modules feature state saving and recovery capabilities, allowing for rapid restoration to the most recent valid state after a system interruption, ensuring the continuity and reliability of the grasping task. Real-time data streams employ a timestamp synchronization mechanism to ensure alignment between force sensor data and visual data in the time dimension. Data is processed using Kalman filtering before fusion to reduce the impact of measurement noise. The system performance monitoring module continuously tracks the utilization of computational resources, dynamically adjusting algorithm parameters to ensure real-time performance requirements. When the computational load is too high, it automatically switches to a simplified computation mode.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A precise positioning and assurance system for intelligent analysis of robotic arm grasping, characterized in that, The system includes: The target recognition module is used to collect surface contour data of the target to be grasped, divide local regions according to the distribution of concave and convex features in the surface contour data, and perform cluster analysis on the surface contour data based on the geometric feature differences of the local regions to obtain each candidate grasping region. The grasping stability assessment module is used to calculate the surface adhesion index of each candidate grasping region based on the geometric feature change trend of the candidate grasping region and the height difference fluctuation between adjacent regions; and to obtain the pressure attenuation difference and pressure distribution difference based on the pressure distribution gradient change trend of the center point of the candidate grasping region and the distance change trend between pressure extreme points. The dynamic interference suppression module is used to generate dynamic interference coefficients for the grasping environment by combining pressure attenuation differences, pressure distribution differences, the average surface adhesion index of all candidate grasping areas, and the average morphological similarity between candidate grasping areas; it constructs the covariance matrix of surface contour data, obtains the eigenvalue sequence through matrix decomposition, and calculates the principal component contribution rate and feature dispersion index based on the concentration and dispersion of the eigenvalues. The positioning compensation module is used to generate precision compensation parameters for grasping and positioning based on the inter-row correlation of the covariance matrix, dynamic interference coefficient, principal component contribution rate, and feature dispersion index; and to perform trajectory correction on the end effector of the robot based on the precision compensation parameters.

2. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 1, characterized in that, The method for obtaining each candidate crawling region is as follows: An edge detection algorithm is used to extract all convex and concave points in the surface contour data. The surface contour data is divided into several sub-regions with each concave point as the dividing boundary. The variance of the geometric features in each sub-region is calculated, and the variance sequence is input into a clustering algorithm to obtain a classification cluster set. The cluster with the largest variance of geometric features within the classification cluster is selected as the preferred cluster, and the sub-regions corresponding to the preferred clusters are marked as candidate crawling regions.

3. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 1, characterized in that, The method for calculating the surface adhesion index of each candidate grasping region is as follows: The region symmetry coefficient and the contact surface adaptation coefficient are calculated based on the fluctuation amplitude of local geometric features in the candidate grasping region and the cumulative change of the height difference between adjacent regions. The product of the regional symmetry coefficient and the contact surface adaptation coefficient is used as the surface adhesion index of the candidate gripping region.

4. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 3, characterized in that, The method for calculating the regional symmetry coefficient and the contact surface adaptation coefficient is as follows: For each candidate capture area, the highest point of the area and its continuous contour points to the left of the area are extracted to form the left contour segment, and the continuous contour points to the right of the highest point are formed the right contour segment. Calculate the first-order difference sequences of the left and right contour segments respectively, and generate a symbolic difference sequence by processing with a symbolic function; The ratio of the absolute value of the cumulative symbolic difference sequence of the left contour segment to the length of the left contour segment is used as the left fitness. The right fitness is calculated using the same method. The harmonic mean of the left fitness and the right fitness is taken as the contact surface fitness coefficient.

5. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 2, characterized in that, The method for obtaining the pressure distribution difference is as follows: The pressure values ​​of the center points of all candidate grabbing areas are sorted by spatial location to generate a pressure sequence, and the coordinates of each pressure extreme point are sorted by distance to generate a location sequence. Calculate the first-order difference sequences of the pressure sequence and the position sequence respectively. Use the variance of the pressure sequence difference sequence as the pressure attenuation difference and the variance of the position sequence difference sequence as the pressure distribution difference.

6. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 1, characterized in that, The method for generating the dynamic interference coefficient is as follows: The arithmetic mean of the surface adhesion index of all candidate gripping regions is calculated as the global adhesion benchmark. The mean Euclidean distance similarity between each candidate capture region and other regions is used as the region morphological similarity. The weighted sum of pressure attenuation difference, pressure distribution difference, global fit benchmark, and mean regional morphological similarity is used as the dynamic interference coefficient.

7. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 1, characterized in that, The method for calculating the principal component contribution rate is as follows: After sorting the feature values ​​in descending order, input them into an adaptive threshold segmentation algorithm to divide the main feature values ​​and secondary feature values. The proportion of the number of principal eigenvalues ​​to the total number of eigenvalues ​​is used as the principal component contribution rate. The product of the mean and variance of the principal eigenvalues ​​is used as the principal eigenvalue volatility, and the product of the mean and variance of the secondary eigenvalues ​​is used as the secondary eigenvalue volatility. The absolute value of the difference between the two is used as the eigenvalue dispersion index.

8. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 1, characterized in that, The method for generating the accuracy compensation parameters is as follows: Calculate the mean absolute value of the cosine similarity between each row of data in the covariance matrix and other rows of data, and use it as the inter-row correlation. The square root of the product of principal component contribution rate, feature dispersion index, inter-row correlation and dynamic interference coefficient is used as the accuracy compensation parameter.

9. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 1, characterized in that, The method for trajectory correction of the robot end effector based on accuracy compensation parameters is as follows: The compensation amount for the joint angle of the robot is generated based on the accuracy compensation parameters. The compensation amount and the accuracy compensation parameters are linearly related in the formula. The compensation amount is converted into the pose adjustment amount of the end effector through inverse kinematics calculation and then superimposed on the original motion trajectory.

10. The precise positioning and assurance system for intelligent analysis of robotic arm grasping as described in claim 1, characterized in that, The system also includes: The real-time feedback module is used to collect contact force data and position deviation data during the actual grasping process after trajectory correction, update the local geometric feature distribution of the surface contour data, and input the updated data into the grasping target recognition module for iterative calculation to form a closed-loop positioning guarantee mechanism.

Images