Robot layered grinding method and device, electronic equipment and medium

Abstract

The invention discloses a robot layered grinding method and device, electronic equipment and a medium, and relates to the technical field of industrial automatic machining.The method comprises the steps that scanning point cloud and target point cloud of a to-be-machined workpiece are obtained, and a global margin field of the surface of the to-be-machined workpiece is obtained through calculation based on normal consistency propagation and directed projection; according to the global margin field and in combination with local geometric features of the workpiece, the self-adaptive cutting depth of each machining point is planned so as to determine the total number of layers; based on the total number of layers and the adaptive cutting depth, discretizing the global margin field in space to generate a plurality of point cloud subsets corresponding to each processing layer; and processing each point cloud subset, extracting and reconstructing structured polishing area data, and polishing according to the structured polishing area data. Under the condition that the surface of the to-be-machined workpiece is rugged, the self-adaptability of polishing layering and the polishing precision are improved.

Description

Technical Field

[0001] This application relates to the field of industrial automated processing technology, and in particular to robotic layer grinding methods, apparatus, electronic devices and media. Background Technology

[0002] In the fields of aerospace, wind turbine blades, and high-end mold manufacturing, surface finishing of carbon fiber composite materials and large castings is a crucial factor determining product performance. Due to the influence of the forming process, these workpieces often exhibit characteristics such as rough surfaces, uneven allowances, and anisotropy. Existing robotic material removal path planning methods mainly fall into two categories: one is offline programming based on ideal CAD models, which cannot adapt to the deformation and errors of actual workpieces caused by casting, clamping, etc., resulting in insufficient machining accuracy. The other is an adaptive planning method based on 3D scanning point clouds, which generates material removal paths by acquiring the actual shape of the workpiece, significantly improving adaptability to actual workpieces.

[0003] Current technologies mostly employ parallel slicing or simple equidistant offset strategies to generate layered paths. This rigid layering logic ignores the differences in local geometric features of the workpiece, which can easily lead to process accidents when processing workpieces such as composite materials that are sensitive to contact forces: in areas with drastic fluctuations in allowance gradients or large curvature, forcibly using a fixed cutting depth can cause excessive instantaneous load on the grinding head, leading to instability of the force control system or even damage to the material; while in areas with flat allowance, a conservative cutting depth setting results in low processing efficiency, making it difficult to meet the requirements of fine grinding.

[0004] Therefore, when the surface of the workpiece to be processed is rugged and the thickness varies, how to improve the adaptability and accuracy of grinding layering is a problem that urgently needs to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a robotic layer-by-layer grinding method, apparatus, electronic device, and medium, which aims to solve the technical problem of how to improve the adaptability and grinding accuracy of layer-by-layer grinding when the surface of the workpiece to be processed is rugged and has different thicknesses.

[0006] To achieve the above objectives, this application proposes a robotic layer-by-layer polishing method, the method comprising: The scanned point cloud and target point cloud of the workpiece to be processed are obtained. Based on the normal uniformity propagation and directional projection, the global margin field of the surface of the workpiece to be processed is calculated. Based on the global margin field and combined with the local geometric features of the workpiece, the adaptive cutting depth of each machining point is planned to determine the total number of layers; Based on the total number of layers and the adaptive cutting depth, the global margin field is discretized in space to generate multiple point cloud subsets corresponding to each machining level; Each of the point cloud subsets is processed to extract and reconstruct structured polishing region data, and polishing is performed based on the structured polishing region data.

[0007] In one embodiment, the step of acquiring the scanned point cloud and target point cloud of the workpiece to be processed, and calculating the global margin field of the surface of the workpiece to be processed based on normal uniformity propagation and directed projection, includes: The scanned point cloud is downsampled using a voxel grid, and a neighborhood point set is searched based on a tree structure. The initial normal vector of each point is estimated through principal component analysis. A Riemann graph connecting the tangent planes of adjacent point clouds is constructed, and a minimum spanning tree strategy is used for traversal propagation. The initial normal vector is propagated in a consistent manner to maximize the dot product of the normal vectors of adjacent points, so as to obtain the propagated normal vector. By combining external viewpoint constraints, all the propagated normal vectors are uniformly redirected to the outside of the surface of the workpiece to be processed, thus obtaining the redirected normal vectors. Based on the redirected normal vector, determine the nearest neighbor of each scan point in the target point cloud, and calculate the directed distance of the current scan point along its normal vector direction to the tangent plane where the nearest neighbor is located; Points with negative directed distances are filtered out, and the remaining positive directed distances form the global margin field.

[0008] In one embodiment, the step of planning the adaptive cutting depth for each machining point based on the global margin field and in conjunction with the local geometric features of the workpiece to determine the total number of layers includes: Statistical analysis is performed on the global margin field, and outliers outside the confidence interval are removed to obtain the global maximum margin. An initial adaptive cutting depth field is constructed based on the local geometric features of the workpiece surface, wherein the local geometric features include the Gaussian curvature and allowance gradient of the workpiece surface; The initial adaptive cutting depth field is smoothed and optimized to obtain the optimized adaptive cutting depth field; The total number of layers is calculated based on the global maximum margin and the reference cutting depth in the optimized adaptive cutting depth field.

[0009] In one embodiment, the step of spatially discretizing the global margin field based on the total number of layers and the adaptive cutting depth to generate multiple point cloud subsets corresponding to each machining level includes: Using the target point cloud as the zero potential energy reference surface, set the iteration index i according to the total number of layers; Based on the optimized adaptive cutting depth field, the spatial mapping distance of the lower reference plane of the current i-th layer is calculated, and the spatial mapping distance of the lower reference plane of the current i-th layer is the cumulative sum of the adaptive cutting depths of the previous i-1 layers; Calculate the spatial mapping distance of the upper reference plane of the current i-th layer, where the spatial mapping distance of the upper reference plane of the current i-th layer is the cumulative sum of the adaptive cutting depth of the previous i layers; Traverse the scanned point cloud, filter out points whose global margin value is greater than the spatial mapping distance of the lower reference plane and less than or equal to the spatial mapping distance of the upper reference plane, and generate the initial point cloud subset of the i-th layer.

[0010] In one embodiment, the step of processing each subset of point clouds to extract and reconstruct structured polishing region data, and then performing polishing based on the structured polishing region data, includes: For each subset of the point cloud, spatial clustering and geometric splitting based on the reference plane are performed sequentially to obtain an optimized cluster of polished sub-regions; Dimensionality reduction and reconstruction are performed on each polishing sub-region cluster to generate structured polishing region data, which includes structured boundary descriptions and their spatial pose information. A grinding path is generated based on the structured boundary description and its spatial pose information, and the workpiece to be processed is ground according to the grinding path.

[0011] In one embodiment, the step of sequentially performing spatial clustering and reference plane-based geometric splitting on each subset of the point cloud to obtain optimized polished sub-region clusters includes: A spatial index structure is constructed based on the initial point cloud subset of the current level, and a density-based spatial clustering algorithm is used to segment it, merging density-connected points into the same cluster to obtain at least one initial cluster. A reference plane is defined by pre-selecting three non-collinear points in space, and the normal vector and distance constant of the current reference plane are calculated. For each initial cluster, calculate the directed distance of all its points relative to the reference plane, and determine whether the points in the current cluster are distributed on both sides of the reference plane. If the distribution is detected on both sides, the current initial cluster is forcibly split into two independent polishing sub-region clusters based on the positive and negative signs of the points on the reference plane. Return to the step of determining whether the points in the current cluster are distributed on both sides of the reference plane, until all processed clusters do not cross the reference plane, thus obtaining the optimized polishing sub-region clusters.

[0012] In one embodiment, the step of performing dimensionality reduction reconstruction on each polishing sub-region cluster to generate structured polishing region data includes: Calculate the three-dimensional coordinate covariance matrix of all points in the current polishing sub-region cluster, and perform eigenvalue decomposition on the three-dimensional coordinate covariance matrix to obtain three eigenvectors, i.e., three principal components; The plane formed by the first two principal components is taken as the best fitting plane for the polishing sub-region cluster, and the two principal components are taken as the basis vectors of the local coordinate system of the current plane, and the centroid of the sub-region cluster is taken as the origin of the local coordinate system. Transform all three-dimensional points in the current polishing sub-region cluster onto the best-fit plane through orthogonal projection to obtain a two-dimensional projection point set; Calculate the convex hull of the two-dimensional projection point set to obtain the minimum convex polygon, which contains all projection points; Based on the minimum convex polygon, the local coordinate system basis vector, and the local coordinate system origin, structured grinding area data is determined, wherein the structured boundary description includes the minimum convex polygon, and the spatial pose information includes the local coordinate system basis vector and the local coordinate system origin.

[0013] Furthermore, to achieve the above objectives, this application also proposes a robotic layer-by-layer polishing device, which includes: The point cloud acquisition module is used to acquire the scanned point cloud and target point cloud of the workpiece to be processed. Based on the normal consistency propagation and directional projection, the global margin field of the surface of the workpiece to be processed is calculated. The layer number determination module is used to plan the adaptive cutting depth of each machining point based on the global margin field and in combination with the local geometric features of the workpiece, so as to determine the total number of layers. The spatial discretization module is used to discretize the global margin field in space based on the total number of layers and the adaptive cutting depth, generating multiple point cloud subsets corresponding to each machining layer. The region reconstruction module is used to process each subset of the point cloud, extract and reconstruct structured polishing region data, and perform polishing based on the structured polishing region data.

[0014] In addition, to achieve the above objectives, this application also proposes an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the robotic layering polishing method described above.

[0015] In addition, to achieve the above objectives, this application also proposes a non-transitory storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the robotic layer-by-layer polishing method described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: The system acquires the scanned point cloud and target point cloud of the workpiece to be processed. Based on normal consistency propagation and directed projection, it calculates the global margin field of the workpiece surface. A high-precision margin calculation method based on normal consistency propagation and directed projection verification is employed, solving the core problem of inaccurate margin assessment caused by point cloud noise and ambiguity in normal direction. Compared with traditional methods that directly calculate distances, it constructs a highly reliable, directionally defined discrete global margin field, laying a precise data foundation for all subsequent intelligent decisions.

[0017] Based on the global allowance field and combined with the local geometric features of the workpiece, an adaptive cutting depth is planned for each machining point to determine the total number of layers. Based on the total number of layers and the adaptive cutting depth, the global allowance field is spatially discretized to generate multiple point cloud subsets corresponding to each machining level. By introducing adaptive cutting depth planning and smooth optimization that incorporates local geometric features, the problem of rigid process parameters that cannot balance force control safety and machining efficiency when dealing with rugged surfaces such as composite materials is solved, achieving a fundamental breakthrough in dynamically and smoothly adjusting the cutting depth according to the workpiece shape. By adopting a multi-level reference surface spatial discretization strategy based on the target surface and the adaptive cutting depth field, the layering logic defect of rigidly dividing continuous and non-uniform allowances is solved, decomposing the complex allowance volume into a series of machining levels with controllable removal amounts and isomorphic to the design model.

[0018] Each point cloud subset is processed to extract and reconstruct structured grinding area data, which is then used for grinding. By comprehensively utilizing data refinement and reconstruction processes, the output is structured grinding area data with clear boundaries and a uniform format, thus providing accurate and effective planning paths for industrial robot grinding.

[0019] For each subset of point cloud, spatial clustering and geometric splitting based on a reference plane are performed sequentially to obtain optimized grinding sub-region clusters. Dimensionality reduction and reconstruction are then performed on each grinding sub-region cluster to generate structured grinding region data, which includes structured boundary descriptions and their spatial pose information. Grinding paths are generated based on these structured boundary descriptions and spatial pose information, and the workpiece to be processed is then ground according to these paths. Because a density-based spatial clustering algorithm is used to process the point cloud subsets in the region segmentation stage, it can more robustly and accurately identify truly independent processing regions based on the actual spatial distribution of the point cloud compared to simple connected component analysis. Furthermore, for complex workpieces with symmetrical structures or deep cavities, a geometric splitting mechanism based on a specified reference plane is introduced. By calculating the directed distance of points relative to the reference plane and forcibly splitting the initial cross-boundary clusters according to the positive and negative signs, the problem of generating chaotic paths in traditional clustering methods is solved, achieving accurate segmentation of processing regions for workpieces with complex topological features. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the robotic polishing system of this application; Figure 2 This is a flowchart illustrating the first embodiment of the robot layer-by-layer polishing method of this application; Figure 3 This is a two-dimensional cross-sectional schematic diagram of the workpiece to be processed being ground according to an embodiment of this application; Figure 4 This is a complete flowchart illustrating the method corresponding to the implementation of this application; Figure 5 This is a schematic diagram of the modular structure of the robot layer grinding device of this application.

[0023] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0025] This application provides a robotic layer-by-layer polishing method, which can be applied to robotic polishing systems. For example... Figure 1 As shown, the robotic polishing system includes an industrial robot, a 3D scanning device, and a host computer. The industrial robot includes a robot body 100 and a polishing end effector 200. The 3D scanning device includes a point cloud acquisition camera 300. The host computer is communicatively connected to both the industrial robot and the 3D scanning device. The host computer stores a computer program, which, when executed by a processor, implements the following robotic layer-by-layer polishing method. Specifically, refer to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the robot layer-by-layer polishing method of this application. In this embodiment, the robot layer-by-layer polishing method includes steps S10 to S40:

[0026] Step S10: Obtain the scanned point cloud and target point cloud of the workpiece to be processed. Based on the normal consistency propagation and directional projection, calculate the global margin field of the surface of the workpiece to be processed.

[0027] It should be noted that the scanned point cloud refers to a series of discrete 3D points in space, acquired by a 3D scanning device, describing the current true shape of the workpiece. The target point cloud refers to a series of discrete 3D points in space, describing the final ideal shape of the workpiece, which can originate from the workpiece's CAD design model. Normal consistency propagation is used to solve the problem of randomness in the direction estimation of normals at single points in the point cloud. By constructing a connection graph between the tangent planes of the point cloud and using a minimum spanning tree strategy to propagate the normal direction, the normal direction of adjacent regions remains continuous and consistent, ultimately being uniformly redirected to the outside of the workpiece. Directed projection calculates the shortest normal distance from a point in the scanned point cloud to the surface defined by the target point cloud. The global margin field can be understood as a set of effective, directed normal projection distance values ​​for all points on the workpiece surface, physically representing the material thickness that needs to be removed at each point on the workpiece surface.

[0028] For example, a point cloud acquisition camera can be used to acquire real 3D scan point cloud data of the workpiece to be processed. Point cloud of the ideal digital model of the workpiece And the two are aligned in the same coordinate system using a point cloud registration algorithm. The scanned point cloud is a discrete set of points acquired by a 3D scanning device that describes the current true shape of a workpiece. The target point cloud is a discrete set of points that describes the final ideal shape of the workpiece, and it is usually derived from the CAD design model of the workpiece.

[0029] Step S20: Based on the global margin field and combined with the local geometric features of the workpiece, plan the adaptive cutting depth for each machining point to determine the total number of layers.

[0030] It should be noted that local geometric features mainly refer to the microscopic geometric properties of the workpiece surface, used to plan the Gaussian curvature and allowance gradient of process parameters. Adaptive depth of cut can be understood as a function that dynamically changes based on the local geometric features of the workpiece surface. After obtaining the robustly estimated global maximum allowance and the optimized adaptive reference depth of cut, the minimum number of machining layers required to remove all the allowance is calculated as the total number of layers.

[0031] Step S30: Based on the total number of layers and the adaptive cutting depth, the global allowance field is discretized in space to generate multiple point cloud subsets corresponding to each machining layer.

[0032] It should be noted that spatial discretization refers to not slicing linearly at fixed intervals, but rather using the ideal target model surface as the zero reference surface. In its normal space, based on a dynamically changing adaptive cutting depth field, a set of virtual, non-uniformly spaced reference surfaces is constructed, thereby decomposing the continuous global margin field values ​​into multiple intervals. The scanned point cloud is traversed, and all points whose directed projection distance values ​​strictly fall within a specific spatial discretization interval are selected. The set of these points constitutes a subset of the point cloud, corresponding to the range to be processed in the i-th machining level.

[0033] Step S40: Process each point cloud subset, extract and reconstruct the structured polishing area data, and perform polishing based on the structured polishing area data.

[0034] It should be noted that after model registration and feature mapping, each subset of point clouds is transformed into a structured processing area. Model registration and feature mapping can be understood as accurately registering the scanned point cloud subset with the CAD model of the workpiece. Then, the geometric features corresponding to this region in the CAD model can be directly extracted, such as the boundaries of patches, holes, and bosses. These features are then projected or mapped onto the point cloud, thus directly obtaining the precise structured processing area boundary defined by the CAD model. Structured grinding area data refers to the standardized data format that is easily processed by industrial robots and is ultimately output for each grinding sub-region.

[0035] In this embodiment, a high-precision margin calculation method based on normal consistency propagation and directed projection verification is adopted, solving the core problem of inaccurate margin assessment caused by point cloud noise and ambiguity of normal direction. Compared with the traditional method of directly calculating distance, a highly reliable and directionally determined discrete global margin field is constructed, laying a precise data foundation for all subsequent intelligent decisions. By introducing adaptive cutting depth planning and smoothing optimization that combines local geometric features, the problem of rigid process parameters that cannot balance force control safety and processing efficiency when dealing with rugged surfaces such as composite materials is solved, achieving a fundamental breakthrough in dynamically and smoothly adjusting the cutting depth according to the workpiece shape. By adopting a multi-level reference surface spatial discretization strategy based on the target surface and the adaptive cutting depth field, the hierarchical logic defect of rigidly dividing continuous and non-uniform margins is solved, decomposing the complex margin volume into a series of processing levels with controllable removal amounts and isomorphic to the design model. By comprehensively utilizing the data refinement and reconstruction process of spatial clustering and geometric split detection, the output is structured grinding area data with clear boundaries and uniform format, thus providing accurate and effective planning paths for industrial robot grinding.

[0036] In one implementation, step S10 includes: Step S101: Voxel mesh downsampling is performed on the scanned point cloud, and a neighborhood point set is searched based on the tree structure. The initial normal vector of each point is estimated through principal component analysis.

[0037] Specifically, the 3D space can be divided into a regular grid of small cubes. Each voxel is represented by its centroid or center point, thus reducing point cloud density, unifying its distribution, and filtering out subtle noise while maintaining the overall geometry. Furthermore, a Kd-tree spatial index data structure is used to quickly find the K nearest neighbors of any given point in space. This efficiently finds the set of points within the local neighborhood of each point, forming the basis for subsequent calculations. Finally, a covariance matrix is ​​constructed from the local neighborhood of a point, and eigenvalue decomposition is performed. The direction of the eigenvector with the smallest eigenvalue is considered perpendicular to the tangent plane at that point, i.e., the estimated initial normal vector.

[0038] For example, the original scanned point cloud can be... Voxel mesh downsampling is performed to preserve the macroscopic geometric features of the workpiece while unifying the surface point cloud density and filtering out high-frequency noise. Subsequently, a K-Nearest Neighbor (KNN) search is performed on each downsampled point based on a Kd-tree spatial indexing structure. The local covariance matrix is ​​constructed using the searched neighborhood point set, and the initial normal vector of the point is calculated through Principal Component Analysis (PCA).

[0039] Step S102: Construct a Riemann graph connecting the tangent planes of adjacent point clouds, and use the minimum spanning tree strategy to perform traversal propagation. Based on maximizing the dot product of the normal vectors of adjacent points, perform consistent propagation on the initial normal vector to obtain the propagated normal vector.

[0040] It's important to note that a Riemannian graph is a graph structure used to describe the local geometric relationships of a point cloud surface. Nodes in the graph are the points in the point cloud, and edges connect adjacent points. The weights of the edges can reflect the differences between the tangent planes or normal directions of two points. The Minimum Spanning Tree (MST) strategy is used to find a subset of all edges in a connected weighted graph that keeps the graph connected while minimizing the total weight of all edges. This strategy is used to determine an optimal path traversing all points in the Riemannian graph, allowing propagation along this path and unifying the normal direction, thus addressing the problem of directional randomness.

[0041] For example, a Riemannian graph connecting the tangent planes of adjacent point clouds can be constructed first, and then traversal propagation can be performed using the MST strategy. The propagation criterion is to maximize the dot product of the normal vectors of adjacent points, i.e. This forces the normal directions within the local neighborhood to remain continuous and consistent.

[0042] Step S103: Combined with external viewpoint constraints, all propagated normal vectors are uniformly redirected to the outside of the surface of the workpiece to be processed, thus obtaining the redirected normal vectors.

[0043] It should be noted that the external viewpoint constraint is used as a reference condition to ultimately determine the uniform pointing direction of all normals. For example, an observation point located outside the workpiece can be specified, such as the position of a scanning device, and all normals, after uniform propagation, can be adjusted to point in the opposite direction to that viewpoint, i.e., pointing outwards from the workpiece surface, thus obtaining a globally uniform normal direction field. For example, by combining the external viewpoint constraint or a preset reference direction, all normals can be uniformly redirected to the outer side of the workpiece surface.

[0044] Step S104: Based on the redirected normal vector, determine the nearest neighbor of each scan point in the target point cloud, and calculate the directed distance from the current scan point along its normal vector direction to the tangent plane where the nearest neighbor is located.

[0045] It should be noted that the directed distance refers to the shortest normal distance from the scan point to the target point cloud surface, and this distance has a direction, which is determined by the redirected normal vector at the scan point. The physical meaning of this distance is the amount of normal deviation from the current scan position to the ideal design surface. A positive value indicates that there is excess material, that is, there is a margin, while a negative value indicates that there is insufficient material, that is, there is overcut or depression.

[0046] For example, each scan point is calculated based on a highly consistent normal field. To the target model The directed normal projection distance of the surface. A normal angle constraint is introduced during this process to calculate the normal vector of the scan point. Its nearest neighbor normal vector on the target model dot product Only when the dot product is greater than a preset threshold, for example... Only when the angle between the two is an acute angle is it considered a valid correspondence, thus eliminating back-to-back mismatches.

[0047] Step S105: Filter out points with negative directed distances, and form a global residual field from the remaining positive directed distances.

[0048] It should be noted that the margin value of points with negative results in the directed distance calculation can be forcibly set to zero, and they will not participate in the subsequent grinding amount calculation. The global margin field is a set consisting of the positive directed distance values ​​at all scan points after filtering.

[0049] For example, the calculated directed distance Perform a sign check. If If the scan point is located inside the surface of the target model, it is determined to be an undercut or overcut region, and the remaining values ​​are set to 0; only the values ​​of the scan point are retained. The positive values ​​are taken as the remaining amount to be removed. Finally, all distance values ​​after the above processing... Together, they form a global residual field, represented as:

[0050] in, The scanned point cloud is a discrete set of points acquired by a 3D scanning device that describes the current true shape of a workpiece. The target point cloud is a discrete set of points that describes the final ideal shape of the workpiece, and it comes from the CAD design model of the workpiece. Represents the target point cloud any point on The unit outward normal vector at point [location] indicates the location of the point [location]. The orientation of the small curved surface; Describe a scalar function that is computed and returns a value. any point in arrive The shortest normal distance of the defined surface, which is physically the distance between points. The amount of material to be processed at that location.

[0051] In this implementation, voxel mesh downsampling and PCA-based initial normal prediction are employed. This first addresses the problem of uneven density and noise in the original scanned point cloud, which leads to instability in local geometric feature calculations, providing a clean and well-organized data foundation for subsequent processing. By introducing a normal consistency propagation strategy based on Riemann diagrams and minimum spanning trees, the inherent problem of random normal direction flipping in local methods like PCA is solved, achieving continuous and smooth normal directions across the entire complex surface. Combined with external viewpoint constraints, the propagated normals are uniformly redirected, resolving the global orientation problem that normals should point outwards from the surface, providing a directional reference for accurate calculation of directed distances. Directed distance calculations are performed based on the aforementioned normal field, and a normal angle verification is introduced, resolving the inaccuracy in margin assessment caused by point cloud registration errors or mismatches between back-to-back points in traditional nearest-point distance calculations, achieving high-fidelity, signed measurement of the actual grinding margin. Setting the margin in overcut regions to zero solves the problem of negative margin interference with subsequent layer planning caused by measurement errors or actual concavities.

[0052] In one implementation, step S20 includes: Step S201: Perform statistical analysis on the global margin field, remove outliers outside the confidence interval, and then process to obtain the global maximum margin.

[0053] It should be noted that, to address the issue of potentially inflated maximum values ​​due to measurement noise in the original residual field D, this implementation method abandons the traditional approach of directly traversing extreme values ​​and instead employs a noise-resistant estimation strategy based on histogram statistics. For example, firstly, a histogram of the numerical distribution of the residual field D is constructed, and the mean μ and standard deviation σ of the residual values ​​are calculated. A confidence interval coefficient k is set, with k=3, and the interval... Maximum values ​​outside the range are identified as outliers and removed. The maximum value in the processed set of remaining values ​​is taken as the effective global maximum remaining value. The corresponding formula is:

[0054] Step S202: Based on the local geometric features of the workpiece surface, an initial adaptive cutting depth field is constructed. The local geometric features include the Gaussian curvature and allowance gradient of the workpiece surface.

[0055] For example, local geometric features include Gaussian curvature K and residual gradient. The adaptive depth-of-cut control function can be defined as follows: Set a larger reference depth of cut in flat areas The reference cutting depth is determined based on the robot's force-position coupling stiffness limit; in regions with high curvature, an attenuation coefficient is introduced to automatically reduce the cutting depth, and the corresponding formula for its calculation logic is as follows: in, The attenuation coefficient is... , To influence the weighting coefficients.

[0056] Step S203: Smooth the initial adaptive cutting depth field to obtain the optimized adaptive cutting depth field.

[0057] For example, to prevent robot oscillation caused by abrupt changes in cutting depth, the initial adaptive cutting depth field can be smoothed using the Laplace operator to obtain an optimized adaptive cutting depth field. .

[0058] Step S204: Calculate the total number of layers based on the global maximum allowance and the reference cutting depth in the optimized adaptive cutting depth field.

[0059] For example, based on optimized baseline depth of cut Compared with statistically corrected Total number of layers The calculation formula is: in, The minimum number of layers required to ensure stable force control throughout the entire process. In this embodiment, a robust estimation of the global maximum allowance is achieved using statistical analysis based on confidence intervals. This effectively solves the core problem of outlier points misleading the total number of layers planning due to scanning noise or registration errors, enabling accurate and stable assessment of the total machining volume. Secondly, local geometric features such as Gaussian curvature and allowance gradient are introduced as control variables into the construction of the adaptive cutting depth field. This solves the problem of rigid process parameters that traditional fixed cutting depths cannot handle uneven surfaces and uneven allowance distributions on composite materials and other workpieces. It allows for dynamic and precise adjustment of the cutting depth based on the workpiece morphology. Furthermore, the initial adaptive field is smoothed using the Laplacian operator, resolving the issue of fluctuations in the cutting depth field caused by abrupt changes in local geometric features, which could lead to force-controlled oscillations and instability during robot machining. This results in a continuously changing and smooth final cutting depth field, laying the foundation for stable machining. Finally, the total number of layers is calculated based on the reference cutting depth in the aforementioned smoothed and optimized cutting depth field and the robustly estimated global maximum allowance, planning the machining process with the theoretically minimum number of layers.

[0060] Based on the above implementation method, step S30 includes: Step S301: Using the target point cloud as the zero potential energy reference surface, set the iteration index i according to the total number of layers.

[0061] It should be noted that the zero potential energy reference surface is a reference surface used for distance measurement; the iteration index is an integer variable used for loop control.

[0062] For example, by constructing a spatial mapping relationship, a continuous residual field can be transformed into a series of logically consistent processing levels. Let the iteration index be... Its value range is An anisotropic space discretization method based on the target surface can be used. Specifically, the algorithm uses an ideal target model. As a zero potential energy reference surface, a set of virtual reference surfaces is constructed in the normal space of the residual field. The spatial spacing of the reference surface is determined by the adaptive cutting depth function. It is determined dynamically.

[0063] Step S302: Calculate the spatial mapping distance of the lower reference plane of the current i-th layer based on the optimized adaptive cutting depth field. The spatial mapping distance of the lower reference plane of the current i-th layer is the cumulative sum of the adaptive cutting depths of the previous i-1 layers.

[0064] It should be noted that the spatial mapping distance of the lower reference plane can be understood as a scalar function defined on each point of the workpiece surface, representing the cumulative thickness of material that has been planned to be removed at that point before the start of the i-th layer of processing.

[0065] For example, for the i-th machining level, its geometric boundary is defined by two adjacent reference surfaces, namely the lower reference surface and the upper reference surface. The lower reference surface corresponds to the starting machining position of the previous layer, i.e., the ending surface of the previous layer, and its corresponding spatial mapping distance, i.e., the inner boundary, is denoted as . The corresponding formula is:

[0066] Step S303: Calculate the spatial mapping distance of the upper reference plane of the current i-th layer. The spatial mapping distance of the upper reference plane of the current i-th layer is the cumulative sum of the adaptive cutting depth of the previous i layers.

[0067] It should be noted that the upper reference plane spatial mapping distance can be understood as a scalar function defined on every point on the workpiece surface, representing the total thickness of material removed at that point after the i-th layer is planned to be processed.

[0068] For example, the upper reference surface corresponds to the processing termination position of the previous layer, and its corresponding spatial mapping distance, i.e., the outer boundary, is denoted as . The corresponding formula is: Step S304: Traverse the scanned point cloud, filter out points whose global margin value is greater than the spatial mapping distance of the lower reference plane and less than or equal to the spatial mapping distance of the upper reference plane, and generate the initial point cloud subset of the i-th layer.

[0069] For example, the screening process can be represented as: Among them, the spatial mapping distance of the lower reference plane and the spatial mapping distance of the upper reference plane together establish the effective range of the current i-th layer in the global range field. Then, the original scanned point cloud was traversed. Based on the directed normal distance value Perform Boolean logic checks to directly filter out all scalar values ​​in the residual field D. Strictly falls within the current range Points within the range, generating an initial point cloud subset. .

[0070] In this embodiment, a multi-level virtual reference surface is dynamically constructed based on the adaptive cutting depth field and the target model surface as the zero potential energy reference surface. This solves the problems of poor geometric fit between the machining trajectory and the complex surface of the workpiece, and the easy distortion in steep or high curvature regions. This ensures that each machining level maintains strict isomorphism with the final design model geometrically. Furthermore, by calculating the spatial accumulation from the previous layer to the current layer using the adaptive cutting depth field, the lower and upper reference surfaces of each layer are defined. This allows for precise customization of the removal amount at different spatial locations for each layer based on local topography and process constraints. Finally, by traversing the original scanned point cloud and using its precise global margin value, all points falling within the dynamic numerical range defined by the upper and lower reference surfaces are rigorously selected to generate the initial point cloud subset for each layer. This accurately decomposes the macroscopic machining task into microscopic operation units that are adaptively planned in space, providing precise input data for the subsequent generation of the robot grinding path.

[0071] In one implementation, step S40 includes: Step S401: For each subset of point clouds, perform spatial clustering and geometric splitting based on the reference plane in sequence to obtain optimized polishing sub-region clusters.

[0072] It should be noted that the initial point cloud subset Spatially, it may consist of multiple interconnected, separate components. To ensure the independence and geometric consistency of subsequent processing, it is possible to... Point cloud spatial clustering is performed. In addition, conventional clustering may not be sufficient to completely distinguish all process areas for symmetrical structures, deep cavities, or specific functional partitions that may exist on the surface of complex workpieces. Therefore, a geometric splitting mechanism based on a specified reference plane is introduced.

[0073] Specifically, step S401 includes: constructing a spatial index structure based on the initial point cloud subset of the current level, and segmenting it using a density-based spatial clustering algorithm, merging density-connected points into the same cluster to obtain at least one initial cluster; pre-selecting three non-collinear points in space to define a reference plane, and calculating the normal vector and distance constant of the current reference plane; for each initial cluster, calculating the directed distance of all its points relative to the reference plane, and determining whether the points in the current cluster are distributed on both sides of the reference plane; if it is detected that they are distributed on both sides, then according to the positive and negative signs of the points on the reference plane, forcibly splitting the current initial cluster into two independent polishing sub-region clusters; returning to the step of determining whether the points in the current cluster are distributed on both sides of the reference plane, until all processed clusters do not cross the reference plane, and obtaining optimized polishing sub-region clusters.

[0074] For example, any clustering algorithm based on spatial neighborhood features, such as Euclidean clustering or region growing algorithms, can be used to segment point clouds. The density-based spatial clustering algorithm DBSCAN can be used to achieve this. Specifically, firstly, a point set... Construct a kd-tree spatial index structure; then define the neighborhood point set. The corresponding formula is:

[0075] Among them, point set It includes all distances to point p within the search radius. Points within the neighborhood. If the number of points in the neighborhood set satisfies , If a threshold is defined, that point is defined as the core point, and all density-connected points are merged into the same cluster using density reachability relationships. Each Each represents an independent cluster of polishing sub-regions. Then, geometric splitting based on a reference plane is performed. Specifically, a reference plane can be pre-selected by choosing three non-collinear points in space, and the normal vector and distance constant of this plane are calculated. For each of the above clusters... The directed distances of all points within a cluster relative to a reference plane are calculated. If points within the cluster are detected to be distributed on both sides of the plane (i.e., the standard deviation of the distance values ​​is greater than a set threshold), the cross-boundary cluster is forcibly split into two independent sub-regions based on the sign of the points on the reference plane. This implementation effectively avoids the generation of grinding trajectories within unreasonable geometric jump ranges, ensuring the process rationality of the sub-regions.

[0076] Step S402: Perform dimensionality reduction and reconstruction on each polishing sub-region cluster to generate structured polishing region data. The structured polishing region data includes structured boundary descriptions and their spatial pose information.

[0077] It should be noted that the 3D point cloud boundary extraction and dimensionality reduction reconstruction algorithm aims to transform discrete point cloud clusters into vectorized polygonal boundaries that are easily processed by computers. This can be achieved using principal component analysis (PCA) combined with a 2D convex hull method. Unlike the local PCA used to calculate the normals of a single point in the previous steps, this step applies PCA to the entire sub-cluster. PCA analysis was performed on the overall point cloud distribution.

[0078] Specifically, step S402 includes: calculating the three-dimensional coordinate covariance matrix of all points in the current grinding sub-region cluster, and performing eigenvalue decomposition on the three-dimensional coordinate covariance matrix to obtain three eigenvectors, i.e., three principal components; taking the plane formed by the first two principal components as the best-fit plane for the grinding sub-region cluster, and taking the two principal components as the local coordinate system basis vectors of the current plane, and taking the centroid of the sub-region cluster as the origin of the local coordinate system; transforming all three-dimensional points in the current grinding sub-region cluster onto the best-fit plane through orthogonal projection to obtain a two-dimensional projection point set; calculating the convex hull of the two-dimensional projection point set to obtain the minimum convex polygon, which contains all projection points; determining the structured grinding region data based on the minimum convex polygon, the local coordinate system basis vectors, and the local coordinate system origin, wherein the structured boundary description includes the minimum convex polygon, and the spatial pose information includes the local coordinate system basis vectors and the local coordinate system origin.

[0079] For example, subclusters can be... Calculate the covariance matrix of the three-dimensional coordinates of all points. The corresponding formula is: in, This is the centroid of the sub-region. Further, the covariance matrix is ​​decomposed into eigenvalues, with the corresponding formula:

[0080] Three feature vectors can be obtained. The three principal components form a locally orthogonal coordinate system for this point cloud cluster. For an approximately planar point cloud cluster, the first two principal components... This forms the best-fit plane for the cluster. Then, all 3D points in this sub-region... By orthogonal projection, transform to... On the defined two-dimensional plane, we obtain a two-dimensional point set. Its projection transformation formula is as follows:

[0081] Finally, these two-dimensional projection points Calculate its two-dimensional convex hull The corresponding formula is: Here, the two-dimensional convex hull is a minimal convex polygon that can contain all projected points, forming a compact, accurate, and easily tractable mathematical description of the boundary of the current grinding sub-region. This implementation reconstructs discrete point cloud data into a structured two-dimensional polygon boundary.

[0082] Step S403: Generate a grinding path based on the structured boundary description and its spatial pose information, and grind the workpiece to be processed according to the grinding path.

[0083] For example, each refined polishing sub-cluster in the aforementioned steps Calculated 2D convex hull polygon Together with its local coordinate system (by of , , (definition), as the first The final structured data for each polishing level is output. After completing the first polishing level... After outputting the data for all structured polishing regions of the layer, the iteration counter... Increment by 1, then return to step S301, and repeat until all... After all levels of data have been generated, the corresponding two-dimensional cross-sectional diagram of this method is shown below. Figure 3 As shown, the complete method flowchart is as follows: Figure 4 As shown.

[0084] In this embodiment, a density-based spatial clustering algorithm is used to process point cloud subsets during region segmentation. Compared with simple connected component analysis, this approach more robustly and accurately identifies truly independent processing regions based on the actual spatial distribution of the point cloud. Furthermore, for complex workpieces with symmetrical structures or deep cavities, a geometric splitting mechanism based on a specified reference plane is introduced. By calculating the directed distance of points relative to the reference plane and forcibly splitting cross-boundary initial clusters according to the positive and negative signs, the problem of generating chaotic paths in traditional clustering methods is solved, achieving accurate segmentation of processing regions for workpieces with complex topological features. Subsequently, by performing dimensionality reduction and convex hull reconstruction based on principal component analysis on each independent grinding sub-region cluster, the discrete and disordered 3D point cloud clusters are transformed. By calculating their covariance matrix, performing eigenvalue decomposition to find the best-fitting plane and local coordinate system, and orthogonally projecting, the 2D convex hull is calculated. This solves the problem of unstructured point cloud data being difficult to directly use for robot path planning, generating a structured description with clear boundaries, compact data, and accurate 3D spatial pose, providing accurate input data for subsequent path planning. Ultimately, the grinding path was generated based on data containing structured boundaries and clear spatial poses, which solved the problem of low path planning quality caused by blurred regional boundaries and missing pose information in traditional methods, and achieved a high degree of fit between the robot grinding trajectory and the geometric features of the workpiece.

[0085] Based on the same inventive concept, such as Figure 5 As shown, this application also provides a robotic layer-by-layer polishing device, which includes: The point cloud acquisition module 10 is used to acquire the scanned point cloud and target point cloud of the workpiece to be processed. Based on the normal consistency propagation and directional projection, the global margin field of the surface of the workpiece to be processed is calculated. The layer number determination module 20 is used to plan the adaptive cutting depth of each machining point based on the global margin field and in combination with the local geometric features of the workpiece, so as to determine the total number of layers. The spatial discretization module 30 is used to discretize the global margin field in space based on the total number of layers and the adaptive cutting depth, generating multiple point cloud subsets corresponding to each machining layer. The region reconstruction module 40 is used to process each subset of the point cloud, extract and reconstruct structured polishing region data, and perform polishing based on the structured polishing region data.

[0086] It should be noted that the robot layer-by-layer polishing device provided in this application embodiment and the robot layer-by-layer polishing method provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned robot layer-by-layer polishing method, and the repeated parts will not be described again.

[0087] In some embodiments, an electronic device provided in this application includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the above-described robotic layered polishing method.

[0088] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0089] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0090] This application also provides a non-transitory storage medium storing a computer program that, when executed by a processor, implements the aforementioned robotic layered polishing method. This storage medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned non-transitory storage medium carries one or more programs, which, when executed, implement the method as described in the embodiments or implementations of this application.

[0091] According to embodiments of this application, a non-transitory storage medium can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. A computer-readable signal medium can also be any storage medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0092] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0093] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A robotic layer-by-layer polishing method, characterized in that, The method includes: The scanned point cloud and target point cloud of the workpiece to be processed are obtained. Based on the normal uniformity propagation and directional projection, the global margin field of the surface of the workpiece to be processed is calculated. Based on the global margin field and combined with the local geometric features of the workpiece, the adaptive cutting depth of each machining point is planned to determine the total number of layers; Based on the total number of layers and the adaptive cutting depth, the global margin field is discretized in space to generate multiple point cloud subsets corresponding to each machining level; Each of the point cloud subsets is processed to extract and reconstruct structured polishing region data, and polishing is performed based on the structured polishing region data.

2. The method as described in claim 1, characterized in that, The steps of acquiring the scanned point cloud and target point cloud of the workpiece to be processed, and calculating the global margin field of the surface of the workpiece to be processed based on normal uniformity propagation and directed projection, include: The scanned point cloud is downsampled using a voxel grid, and a neighborhood point set is searched based on a tree structure. The initial normal vector of each point is estimated through principal component analysis. A Riemann graph connecting the tangent planes of adjacent point clouds is constructed, and a minimum spanning tree strategy is used for traversal propagation. The initial normal vector is propagated in a consistent manner to maximize the dot product of the normal vectors of adjacent points, so as to obtain the propagated normal vector. By combining external viewpoint constraints, all the propagated normal vectors are uniformly redirected to the outside of the surface of the workpiece to be processed, thus obtaining the redirected normal vectors. Based on the redirected normal vector, determine the nearest neighbor of each scan point in the target point cloud, and calculate the directed distance of the current scan point along its normal vector direction to the tangent plane where the nearest neighbor is located; Points with negative directed distances are filtered out, and the remaining positive directed distances form the global margin field.

3. The method as described in claim 1, characterized in that, The step of planning the adaptive cutting depth for each machining point based on the global margin field and in combination with the local geometric features of the workpiece to determine the total number of layers includes: Statistical analysis is performed on the global margin field, and outliers outside the confidence interval are removed to obtain the global maximum margin. An initial adaptive cutting depth field is constructed based on the local geometric features of the workpiece surface, wherein the local geometric features include the Gaussian curvature and allowance gradient of the workpiece surface; The initial adaptive cutting depth field is smoothed and optimized to obtain the optimized adaptive cutting depth field; The total number of layers is calculated based on the global maximum margin and the reference cutting depth in the optimized adaptive cutting depth field.

4. The method as described in claim 3, characterized in that, The step of discretizing the global margin field spatially based on the total number of layers and the adaptive cutting depth to generate multiple point cloud subsets corresponding to each machining level includes: Using the target point cloud as the zero potential energy reference surface, set the iteration index i according to the total number of layers; Based on the optimized adaptive cutting depth field, the spatial mapping distance of the lower reference plane of the current i-th layer is calculated, and the spatial mapping distance of the lower reference plane of the current i-th layer is the cumulative sum of the adaptive cutting depths of the previous i-1 layers; Calculate the spatial mapping distance of the upper reference plane of the current i-th layer, where the spatial mapping distance of the upper reference plane of the current i-th layer is the cumulative sum of the adaptive cutting depth of the previous i layers; Traverse the scanned point cloud, filter out points whose global margin value is greater than the spatial mapping distance of the lower reference plane and less than or equal to the spatial mapping distance of the upper reference plane, and generate the initial point cloud subset of the i-th layer.

5. The method as described in claim 1, characterized in that, The steps of processing each subset of point clouds, extracting and reconstructing structured polishing region data, and performing polishing based on the structured polishing region data include: For each subset of the point cloud, spatial clustering and geometric splitting based on the reference plane are performed sequentially to obtain an optimized cluster of polished sub-regions; Dimensionality reduction and reconstruction are performed on each polishing sub-region cluster to generate structured polishing region data, which includes structured boundary descriptions and their spatial pose information. A grinding path is generated based on the structured boundary description and its spatial pose information, and the workpiece to be processed is ground according to the grinding path.

6. The method as described in claim 5, characterized in that, The step of sequentially performing spatial clustering and reference plane-based geometric splitting processing on each subset of the point cloud to obtain optimized polished sub-region clusters includes: A spatial index structure is constructed based on the initial point cloud subset of the current level, and a density-based spatial clustering algorithm is used to segment it, merging density-connected points into the same cluster to obtain at least one initial cluster. A reference plane is defined by pre-selecting three non-collinear points in space, and the normal vector and distance constant of the current reference plane are calculated. For each initial cluster, calculate the directed distance of all its points relative to the reference plane, and determine whether the points in the current cluster are distributed on both sides of the reference plane. If the distribution is detected on both sides, the current initial cluster is forcibly split into two independent polishing sub-region clusters based on the positive and negative signs of the points on the reference plane. Return to the step of determining whether the points in the current cluster are distributed on both sides of the reference plane, until all processed clusters do not cross the reference plane, thus obtaining the optimized polishing sub-region clusters.

7. The method as described in claim 5, characterized in that, The step of performing dimensionality reduction and reconstruction on each polishing sub-region cluster to generate structured polishing region data includes: Calculate the three-dimensional coordinate covariance matrix of all points in the current polishing sub-region cluster, and perform eigenvalue decomposition on the three-dimensional coordinate covariance matrix to obtain three eigenvectors, i.e., three principal components; The plane formed by the first two principal components is taken as the best fitting plane for the polishing sub-region cluster, and the two principal components are taken as the basis vectors of the local coordinate system of the current plane, and the centroid of the sub-region cluster is taken as the origin of the local coordinate system. Transform all three-dimensional points in the current polishing sub-region cluster onto the best-fit plane through orthogonal projection to obtain a two-dimensional projection point set; Calculate the convex hull of the two-dimensional projection point set to obtain the minimum convex polygon, which contains all projection points; Based on the minimum convex polygon, the local coordinate system basis vector, and the local coordinate system origin, structured grinding area data is determined, wherein the structured boundary description includes the minimum convex polygon, and the spatial pose information includes the local coordinate system basis vector and the local coordinate system origin.

8. A robotic layer-by-layer polishing device, characterized in that, The robotic layer-by-layer polishing device includes: The point cloud acquisition module is used to acquire the scanned point cloud and target point cloud of the workpiece to be processed. Based on the normal consistency propagation and directional projection, the global margin field of the surface of the workpiece to be processed is calculated. The layer number determination module is used to plan the adaptive cutting depth of each machining point based on the global margin field and the local geometric features of the workpiece, so as to determine the total number of layers. The spatial discretization module is used to discretize the global margin field in space based on the total number of layers and the adaptive cutting depth, generating multiple point cloud subsets corresponding to each machining layer. The region reconstruction module is used to process each subset of the point cloud, extract and reconstruct structured polishing region data, and perform polishing based on the structured polishing region data.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the robotic layered polishing method as described in any one of claims 1 to 7.

10. A non-transitory storage medium, characterized in that, The non-transitory storage medium stores a computer program, which, when executed by a processor, implements the steps of the robot layering polishing method as described in any one of claims 1 to 7.

Images