A method and system for planning a quasi-circular cable grasp

Abstract

The application relates to the technical field of cable grabbing, and discloses a kind of round cable grabbing planning method and system, solve the problem that the existing technology does not provide the grabbing position of cable is prone to cause cable grabbing failure and reduce cable safety.The round cable grabbing planning method provided by the application comprises the following steps: obtaining point cloud data of the cable, classifying based on the point cloud data, sorting based on the classified point cloud data, fitting a circle model based on the sorted point cloud data, segmenting the coil point cloud based on the tangent plane of the circle model, and then calculating the best grabbing position of the cable, thereby ensuring the safety and effectiveness when grabbing the round cable or the circular cable.

Description

Technical Field

[0001] This invention relates to the field of cable gripping technology, specifically to a planning method and system for gripping quasi-circular cables. Background Technology

[0002] With rising labor costs and an increasing pursuit of production efficiency, the structure of the world's manufacturing industry is shifting towards full automation, leading to a surge in demand for automated production. Cable loading and handling are typically done manually, but due to high labor costs and generally low efficiency, traditional methods are no longer sufficient to meet current production demands.

[0003] Therefore, various methods for robots to grasp cables have emerged, such as the "Master-Slave Mechanical Arm Device for Power Field Inspection Robot" with application number 202010695642.3, which realizes the grasping of cables. However, cables are generally round or near-round, and the length of the cables varies. Existing devices do not provide the optimal grasping position for the cables. If the grasping is not performed from the optimal grasping position, the cable grasping may fail. Moreover, the robot's arm or hand is generally made of relatively hard material, which can easily damage the surface of the cable and reduce the safety of the cable. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a planning method and system for grasping quasi-circular cables, which solves the problem that the lack of information on the cable grasping position in existing technologies easily leads to cable grasping failure and reduced cable safety.

[0006] (II) Technical Solution

[0007] To address the above problems, the present invention provides the following technical solution: a planning method for grasping quasi-circular cables, characterized in that: the planning method includes:

[0008] Acquire point cloud data of the cables and classify them based on the point cloud data;

[0009] The sorted point cloud data is then used to fit a circle model to the sorted point cloud data.

[0010] The coil point cloud is segmented based on the tangent plane of the circular model, and then the optimal gripping position of the cable is calculated.

[0011] The present invention also provides the following technical solution: a planning system for grasping quasi-circular cables, the planning system comprising:

[0012] The classification module is used to acquire point cloud data of cables and classify them based on the point cloud data;

[0013] The fitting module is used to sort the classified point cloud data and fit a circle model to the sorted point cloud data.

[0014] The calculation module is used to segment the coil point cloud based on the tangent plane of the circular model, and then calculate the optimal gripping position of the cable.

[0015] (III) Beneficial Effects

[0016] Compared with the prior art, the present invention provides a planning method and system for grasping quasi-circular cables, which has the following beneficial effects:

[0017] 1. The planning method and system first acquires the point cloud data of the cable, then obtains the coil point cloud by classifying the point cloud data, fits the coil point cloud to obtain a circular model, and divides the circular model by the tangent plane of the circular model, thereby calculating the optimal gripping position of the cable, thus realizing the effective gripping of quasi-circular or circular cables, avoiding damage to the outer shell of quasi-circular or circular cables, and ensuring the safety of quasi-circular or circular cables.

[0018] 2. In the process of obtaining a circular model by fitting a coil point cloud, the planning method and system removes data points when the distance between a data point in the point cloud data and the fitted circle is greater than a certain threshold. This process is repeated to ensure the accuracy of the obtained circular model and avoid interference from such data points in the point cloud data where the distance between the data point and the fitted circle is greater than the certain threshold.

[0019] 3. The planning method and system calculate the covariance matrix of the point cloud data based on the centroid coordinates of the segmented blocks, obtain the eigenvalues ​​and eigenvectors of the covariance matrix, and then obtain multiple grasping points. A collision rectangle is established from the multiple grasping points to obtain the optimal grasping point, thereby ensuring the safety and effectiveness when grasping circular or round cables. Attached Figure Description

[0020] Figure 1 A schematic diagram of point cloud data segmented by the AI ​​model in the planning method provided by the present invention;

[0021] Figure 2 This is a schematic diagram of the optimal grab point provided in the planning method of the present invention.

[0022] Figure 3 A flowchart of the planning method for grabbing quasi-circular cables provided by the present invention.

[0023] Figure 4 The structural diagram of the planning system for grasping quasi-circular cables provided by the present invention. Detailed Implementation

[0024] 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.

[0025] Please see Figures 1-3 This invention provides a new technical solution: a planning method for grasping quasi-circular cables, the planning method comprising:

[0026] S1. Obtain the point cloud data of the cable and classify it based on the point cloud data.

[0027] Specifically, the cable is generally circular or near-circular. It is scanned using 3D methods to obtain point cloud data. The resulting point cloud data is substantial. A pre-set AI model is used to segment the point cloud data and classify it into corresponding cable parts, including coils, wire ends, and junction boxes. The pre-set AI model is pre-trained, using a large number of pre-collected 2D cable images as samples. The coils, wire ends, and junction boxes in the images are labeled, trained, and iterated using a neural network algorithm to obtain the AI ​​model. The scanned point cloud data is segmented using the AI ​​model into multiple regions. These segmented regions have a corresponding mapping relationship with the point cloud data; therefore, the cable point cloud data can be classified as coil point clouds, i.e.,... Figure 1 The circular and line-shaped dots in the cloud are Figure 1 A shape with one pointed end and one square end, a line box dot cloud, that is... Figure 1 A shape with squares at both ends.

[0028] Furthermore, the classification based on point cloud data includes:

[0029] Randomly select one data point from the point cloud data as the initial clustering reference point. If the distance between any point cloud data and the initial clustering reference point is less than a preset threshold, then these two point cloud data are classified into the same category.

[0030] For the remaining unclassified point cloud data, select the nearest classified data point as the clustering reference point, and classify the point cloud data that are within a preset threshold distance from the clustering reference point into that class.

[0031] Specifically, the point cloud data of the cable includes multiple data points, each of which is a point in three-dimensional space corresponding to a certain location in the three-dimensional space. A data point is randomly selected as the initial clustering reference point. If the distance between any point cloud data and the initial clustering reference point is less than a preset threshold (let the preset threshold be T), then these two point cloud data are grouped into the same category. For the remaining unclassified point cloud data, the closest classified data point is selected as the clustering reference point. There can be multiple clustering reference points, and point cloud data whose distance to the clustering reference point is within the preset threshold T are grouped into that category, thus obtaining multiple classification sets. From all the classification sets, only the classification set with a certain number of data points in the point cloud data is retained as the final filtering result. Some classification sets with fewer points are deleted to avoid interference from these smaller classification sets to larger classification sets, thereby achieving noise reduction. This classifies the point cloud data into coil point clouds, wire end point clouds, and wire box point clouds, thus achieving the effect of classifying the point cloud data.

[0032] S2. Sort the classified point cloud data and fit a circle model based on the sorted point cloud data.

[0033] Specifically, the classification sets obtained after classification are coil point clouds, wire end point clouds, and wire box point clouds. The coil point clouds are sorted in descending order according to the z value of the data point pi(x,y,z) to obtain the highest point pimax(x,y,z) on each coil point cloud. The same method is used to sort all pimax(x,y,z) in descending order according to the z value, thereby obtaining a coil arrangement from high to low. A circle is then fitted based on the sorted coil point clouds to obtain a circle model.

[0034] Furthermore, fitting the sorted point cloud data into a circle includes:

[0035] Three data points are randomly selected as the starting points of the initial fitting circle. The center and radius of the fitting circle are calculated based on these three data points. All data points are traversed, and the distance from each data point to the fitting circle is calculated. If the distance from all data points to the fitting circle is less than a specific threshold, then this fitting circle is a circular model.

[0036] Specifically, three data points are randomly selected as the starting points of the initial fitted circle. The center and radius of the fitted circle are calculated based on these three data points. All data points are iterated over, and the distance from each data point to the fitted circle is calculated to evaluate the fitting effect of the current fitted circle. If the distance from all data points to the fitted circle is less than a specific threshold, the fitted circle is accurate enough, and the iteration process can be terminated. The fitted circle is then called the circular model, with the following form: (xa)2 + (yb)2 = r2, where (a,b) represents the coordinates of the center of the circle, and r represents the radius of the circle. If any data point is more than a specific threshold in distance from the fitted circle, such data points are removed, and the above process is repeated until the distance from all data points to the fitted circle is less than a specific threshold, at which point the circular model is obtained.

[0037] S3. The coil point cloud is segmented based on the tangent plane of the circular model, and then the optimal gripping position of the cable is calculated.

[0038] Specifically, based on the circular model obtained above, the tangent plane of the coil point cloud represented by the circular model is calculated. Based on the obtained tangent plane, the coil point cloud is segmented to obtain multiple segmented blocks. Principal component analysis is performed on the multiple segmented blocks to obtain multiple grab points of the cable. Rectangular collision detection is performed on the obtained grab points to obtain the optimal zone position from the multiple grab points.

[0039] Furthermore, the step of segmenting the coil point cloud using the tangent plane of the circular model includes:

[0040] Based on the circular model, a planar bundle tangent to the circular model is obtained. The initial plane of the planar bundle is rotated to obtain the planar model at each position. Based on the distance from the data point to the planar model, the point cloud at each position within a preset threshold is calculated to obtain the segmentation block at each position.

[0041] Specifically, let the equation of the planar bundle of the tangent plane tangent to the circular model be: Ax + By + Cz + D = 0. Then let the equation of the initial position plane of the planar bundle be: A1*x + B1*y + C1*z + D1 = 0. The central axis vector of the coil point cloud represented by the circular model is: v(x,y,z). Rotate the initial position plane by 360° with a step size s and vector v to obtain multiple positions. The number of positions is the quotient of 360° and step size s, resulting in the planar model for each position: A i *x+B i *y+C i *z+D i =0, then a tangent plane is calculated at each position. Multiple tangent planes at multiple positions form a planar bundle. Let the distance from each data point in the point cloud data to the planar model at each position be , where the distance is... Set a preset distance t between the point cloud data and the planar model, calculate the point cloud data within the preset distance t at each position, that is, extend the tangent planes at multiple positions to both sides by a certain distance to the preset distance t, and take the coil point cloud within the extension range of each tangent plane as the segment block of each position.

[0042] Furthermore, the step of calculating the optimal gripping position for the cable includes:

[0043] The covariance matrix of the point cloud data is calculated based on the centroid coordinates of the segmented blocks. The eigenvalues ​​and eigenvectors of the covariance matrix are then obtained based on the covariance matrix, and the capture points are obtained.

[0044] Specifically, principal component analysis is performed on each segmented block obtained at each location. First, the centroid coordinates q(x,y,z) are calculated from the point cloud data of the segmented block. That is, each segmented block has only one centroid, and each centroid corresponds to one grasping point. Therefore, multiple segmented blocks have multiple centroids, and thus a coil point cloud has multiple centroids, which in turn has multiple grasping points. All data points are then centered pi(x,y,z) = pi(x,y,z) - q(x,y,z), and the covariance matrix C of the point cloud data is calculated, where C(i,j) = (1 / n) * X'X, which is represented by the i-th dimension. The covariance matrix between the j-th dimension and the segmented block is calculated, where X represents the mn matrix formed by n data points in the segmented block, X' represents the transpose of X, n is the number of data points in the segmented block, and m=3, i.e., three-dimensional space. The eigenvalues ​​and eigenvectors of the covariance matrix C are calculated, and the obtained eigenvectors are sorted according to the size of the corresponding eigenvalues. The first principal component (i.e., the eigenvector corresponding to the largest eigenvalue) is taken as the X direction vx of the spatial direction of the grasping point, the Z direction is taken as vz(0,0,1), and the Y direction is determined according to the right-hand coordinate system vy. Through actual use, it is found that the centroid q is reasonable as the grasping point.

[0045] Furthermore, based on the covariance matrix, the eigenvalues ​​and eigenvectors of the covariance matrix are obtained, and thus the capture points are obtained, including:

[0046] Based on the gripping points, collision rectangles are created in the collision bounding box, the size of the gripping points is calculated, and the optimal gripping position is obtained from them.

[0047] Specifically, let the two-dimensional orthographic size of the gripper of the robot during grasping be S(w, h), where w is the length of the gripper in the X direction of the tool coordinate system, h is the length of the gripper in the Y direction of the tool coordinate system, and a collision rectangle R = {(x1,y1,0)(x2,y2,0)(x3,y3,0)(x4,y4,0)} is established for each grasping point, where x1 = q + w / 2*vx, y1 = q + h / 2*vy, x2 = q + w / 2*vx, y2 = q - h / 2*vy, x3 = q - w / 2*vx, y3 = q - h / 2*vy, x4 = q - w / 2*vx, y4 = q + h / 2*vy. Let the collision bounding box B = {R1,R2}, R1 = {(x1,y1,-1000)(x2,y2,-1000)(x3,y3,-1000)(x4,y4,-1000)}, R2 = {(x1,y1,1000)(x2,y2,1000)(x3,y3,1000)(x4,y4,1000)}. Extract the segmented blocks in the spatial bounding box B, and calculate the dimensions K(wk, hk) in the two directions of vx and vy. If wk < w and hk < h, the grasping point is graspable, and the grasping point with the smallest hk is selected from all graspable points as the optimal grasping point, that is Figure 2 the position indicated by the middle arrow, so as to complete the cable grasping task.

[0048] The present invention further provides a planning system for quasi-circular cable grasping, as shown in Figure 4 , the planning system includes:

[0049] a classification module, configured to acquire point cloud data of the cable and perform classification based on the point cloud data;

[0050] a fitting module, configured to sort the classified point cloud data, and fit the sorted point cloud data into a circle model;

[0051] a calculation module, configured to segment the coil point cloud based on the tangent plane of the circle model, and then calculate to obtain the optimal grasping position of the cable.

[0052] Further, the classification module is specifically configured to:

[0053] randomly select one data point in the point cloud data as the initial clustering reference point, and if the distance between any point cloud data and the initial clustering reference point is less than a preset threshold, classify the two point cloud data into the same category;

[0054] for the remaining unclassified point cloud data, select the nearest classified data point therein as the clustering reference point, and classify the point cloud data whose distance to the clustering reference point is within the preset threshold into this category.

[0055] Further, the fitting module is specifically configured to:

[0056] Three data points are randomly selected as the starting points of the initial fitting circle. The center and radius of the fitting circle are calculated based on these three data points. All data points are traversed, and the distance from each data point to the fitting circle is calculated. If the distance from all data points to the fitting circle is less than a specific threshold, then this fitting circle is a circular model.

[0057] Furthermore, the computing module is specifically used for:

[0058] Based on the circular model, a planar bundle tangent to the circular model is obtained. The initial plane of the planar bundle is rotated to obtain the planar model at each position. Based on the distance from the data point to the planar model, the point cloud at each position within a preset threshold is calculated to obtain the segmentation block at each position.

[0059] Furthermore, the computing module is specifically used for:

[0060] The covariance matrix of the point cloud data is calculated based on the centroid coordinates of the segmented blocks. The eigenvalues ​​and eigenvectors of the covariance matrix are then obtained based on the covariance matrix, and the capture points are obtained.

[0061] Furthermore, the computing module is specifically used for:

[0062] Based on the gripping points, collision rectangles are created in the collision bounding box, the size of the gripping points is calculated, and the optimal gripping position is obtained from them.

[0063] The planning system for capturing quasi-circular cables provided in this application can implement all the processes of the above-described planning method for capturing quasi-circular cables and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0064] 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 planning method for grasping quasi-circular cables, characterized in that: The planning method includes: Obtain point cloud data of circular cables. Randomly select one data point from the point cloud data as the initial clustering reference point. If the distance between any point cloud data and the initial clustering reference point is less than a preset threshold, then these two point cloud data are classified into the same category. For the remaining unclassified point cloud data, select the closest classified data point as the clustering reference point, and classify the point cloud data whose distance from the clustering reference point is within the preset threshold into that category. This completes the point cloud classification, resulting in coil point cloud, wire end point cloud, and wire box point cloud. Based on the classified coil point cloud sorted in descending order of z value, three data points are randomly selected as the starting points of the initial fitted circle. The center and radius of the circle are calculated. All data points are traversed and outliers that are more than a certain threshold away from the fitted circle are removed. The circle model is obtained iteratively. Based on the circular model, a planar bundle tangent to the circular model is obtained. The initial plane of the planar bundle is rotated 360° around the central axis of the circle with a fixed step size to obtain the planar model at each position. The point cloud within a preset distance threshold at each position is extracted to obtain the corresponding segmentation block. The covariance matrix is ​​calculated based on the centroid coordinates of the segmentation block. The grasping point is obtained through the eigenvalues ​​and eigenvectors of the covariance matrix. Based on the grasping point, a collision rectangle is established in the collision bounding box, the size of the grasping point is calculated, and the optimal grasping position is obtained from it.

2. A planning system for grasping quasi-circular cables, characterized in that: The planning system includes: The classification module is used to acquire point cloud data of circular cables. It randomly selects a data point in the point cloud data as the initial clustering reference point. If the distance between any point cloud data and the initial clustering reference point is less than a preset threshold, the two point cloud data are classified into the same category. For the remaining unclassified point cloud data, the closest classified data point is selected as the clustering reference point, and the point cloud data whose distance to the clustering reference point is within the preset threshold are classified into that category. This completes the point cloud classification and yields coil point cloud, wire end point cloud, and wire box point cloud. The fitting module is used to sort the classified coil point cloud in descending order of z-value, randomly select three data points as the starting points of the initial fitting circle, calculate the center and radius of the circle, traverse all data points and remove outliers whose distance to the fitting circle is greater than a certain threshold, and iterate to obtain the circle model. The calculation module is used to obtain a planar bundle tangent to the circular model based on the circular model. The initial plane of the planar bundle is rotated 360° around the central axis of the circle with a fixed step size to obtain the planar model at each position. The point cloud within a preset distance threshold at each position is extracted to obtain the corresponding segmentation block. The covariance matrix is ​​calculated based on the centroid coordinates of the segmentation block. The grasping point is obtained through the eigenvalues ​​and eigenvectors of the covariance matrix. The collision rectangle is established in the collision bounding box based on the grasping point, the size of the grasping point is calculated, and the optimal grasping position is obtained from it.

Images