A robot optimal grasp contact point matching and grasp quality evaluation method and system

Abstract

The application discloses a kind of robot optimal grasping contact point matching and grasping quality evaluation method and system, first based on the three-dimensional point cloud of target object, generate the spherical view vector covering its potential grasping orientation and corresponding grasping approach vector. Then the projection distance of each point to be processed on the surface of object in each spherical view direction is calculated, combined with multidimensional physical constraint, the largest projection distance of stable contact point pair is screened out, and the grasping width is determined accordingly. Then based on the above contact point pair and grasping approach vector, candidate grasping posture is constructed, the robustness of surface smoothness of grasping is evaluated by calculating minimum friction coefficient, collision detection is carried out to ensure operation safety, and whether the grasping width meets the physical limit of gripper is verified. Finally, the grasping posture that meets the standards in mechanical stability, collision-free feasibility and physical realizability is screened out by comprehensively screening the above indexes. Through multidimensional evaluation, robot grasping scheme suitable for complex scene is efficiently and reliably generated.

Description

Technical Field

[0001] This invention belongs to the field of robot target grasping technology, and in particular relates to a method and system for optimal robot grasping contact point matching and grasping quality evaluation. Background Technology

[0002] With the deep integration of industrial automation and intelligent manufacturing, the application of robots in key scenarios such as precision assembly, logistics sorting, and flexible production lines has become a core force for improving production efficiency. As the underlying fundamental capability for robots to achieve physical interaction, stable and reliable grasping operations directly determine the efficiency, robustness, and intelligence level of the entire automation system. Although researchers have conducted extensive research on robot grasping, high-reliability robot grasping tasks still face many challenges due to the diversity of target object shapes, appearances, and sizes, as well as the influence of clutter interference and occlusion phenomena in the working environment.

[0003] In recent years, research on robot grasping has been mainly divided into model-based methods and data-driven methods. Model-based methods usually assume the known 3D model and category of the object, and generate the grasping pose by pre-defining the grasping configuration and combining it with the estimated 6D object pose. However, they are highly dependent on the accurate model, only applicable to known objects, and have limited adaptability to new or irregular objects. They also often require additional collision detection as post-processing. In contrast, data-driven methods do not rely on strong prior models, but learn the grasping strategy directly from visual input, and have better generalization ability. However, when predicting 6-DOF grasping poses in the complete SE(3) space, grasping usually needs to describe multiple parameters simultaneously, resulting in high dimensionality and a large number of parameters in the grasping representation, which significantly increases the modeling and learning complexity of the prediction algorithm. Specifically, regarding key technical aspects, in contact point matching, existing methods mostly rely on relatively simple geometric metrics (such as Euclidean distance or the angle between normal vectors) for selection. While computationally efficient, these methods cannot comprehensively consider complex constraints across multiple dimensions, such as spatial alignment, directional consistency, and force closure. Consequently, the accuracy and robustness of matching drop sharply in occluded or cluttered scenarios. In terms of grasping quality evaluation, existing systems often rely on single indicators (such as focusing only on friction cone conditions or geometric stability), lacking a comprehensive evaluation model that integrates multiple physical constraints. This single-dimensional assessment fails to accurately depict the complex physical interactions during real-world grasping, a significant reason for the high failure rate of generated grasping postures in practical applications.

[0004] In summary, existing technologies have significant shortcomings in terms of the accuracy of contact point matching and the comprehensiveness of grasping quality assessment. Current methods cannot effectively coordinate multiple physical constraints, which severely limits the adaptability and success rate of robotic grasping systems in complex scenarios. Therefore, a new method is urgently needed to improve the accuracy of contact point matching and the comprehensiveness of grasping quality assessment, thereby enhancing the reliability and efficiency of robotic grasping tasks in diverse application scenarios. Summary of the Invention

[0005] This invention proposes a method and system for optimal robot grasping contact point matching and grasping quality evaluation. Addressing the practical difficulties in robot grasping detection, namely (1) insufficient generalization of grasping in complex scenarios and (2) a single evaluation index, this invention (1) provides a multi-dimensional constrained contact point matching method, overcoming the limitations of existing methods that rely solely on geometric features for screening and matching. This significantly improves the robustness and accuracy of contact point matching, thereby ensuring that the robot can achieve efficient and stable grasping on objects with complex geometries and diverse grasping requirements; (2) establishes a complete grasping quality evaluation process, combining multiple key factors such as friction coefficient, grasping width, and scene collision to comprehensively evaluate the physical rationality of the grasping posture, accurately reflecting potential risk factors during the grasping process, and effectively improving the stability and reliability of the grasping operation.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] A method for optimal robot grasping contact point matching and grasping quality evaluation, the method comprising the following steps:

[0008] S100: Obtain the 3D point cloud of the target object, generate a set of spherical view vectors covering the potential orientation of the object to be grasped based on the 3D point cloud, and generate a grasping proximity vector perpendicular to the spherical view vector.

[0009] S200: For each point to be processed on the surface of the object and its corresponding spherical view vector direction, calculate the projection distance of the point to be processed and all other points in the current spherical view vector direction. By applying multiple physical constraints, candidate point pairs are screened. The effective point pair with the largest projection distance is determined as the optimal contact point pair that can form a stable grip under the current view vector direction. The maximum projection distance is recorded. The corresponding gripping width is determined based on the maximum projection distance. Candidate gripping postures are formed based on the optimal contact point pair and the gripping proximity vector.

[0010] S300: Calculates the minimum coefficient of friction required to achieve stable gripping to assess robustness to object surface smoothness, performs collision detection of gripping posture to ensure operational safety, and verifies that the gripping width is within the physical limits of the gripper.

[0011] S400: Based on the minimum friction coefficient, collision detection results, and gripping width, the gripping posture that meets the requirements in terms of mechanical stability, collision-free feasibility, and physical feasibility is selected.

[0012] Preferably, based on the Fibonacci grid sampling algorithm, a set of uniformly distributed spherical view vectors covering the potential orientation of the object to be grasped is generated on a unit sphere. S100 generates the spherical view vectors including:

[0013] S110: Create from 0 to The index sequence, calculate the index of each sampling point. The coordinates on the axis are calculated using trigonometric functions. shaft and The coordinate components of the axes are defined by the following formulas:

[0014] ;

[0015] in, This indicates the total number of viewpoints that need to be generated. For indexing, It is an angular parameter related to the golden ratio. , , For the first The coordinate components of each sampling point on the unit sphere;

[0016] S120: Stack the calculated three coordinate components to form a two-dimensional matrix, and map the points on the unit sphere to the actual sphere through scaling and translation transformations, finally obtaining a complete set of spherical view vectors. Specifically:

[0017] ;

[0018] in, Define the three-dimensional coordinates of the sphere's center, which is located at the origin by default. Set to the radius of the sphere; the default value is the unit length.

[0019] Preferably, generating a grasping proximity vector perpendicular to the spherical viewpoint vector in S100 includes:

[0020] S130: For each spherical view vector Using a preset reference vector An initial grasping proximity vector perpendicular to it is constructed using vector projection. Specifically:

[0021] ;

[0022] in, Represents the multiplication operation between a scalar and a vector;

[0023] S140: When generating the grasp proximity vector, use the Rodriguez rotation formula to make the initial grasp proximity vector... Around the spherical view vector Rotate at preset equal intervals Generate a set of grab proximity vectors that are uniformly distributed in space. The Rodriguez rotation formula is as follows:

[0024] ;

[0025] in, It is the rotation angle. Represents the cross product of vectors. Represents the dot product of vectors.

[0026] Preferably, in step S200, calculating the projection distance includes:

[0027] S210: Spherical View Vector Copy and expand to a dimension matching the number N of object gripping points, as the view direction vector originating from each sampling point. This represents the various possible grabbing directions starting from each grabbing point and surrounding that point;

[0028] S220: For each point to be processed, calculate the vector between the current point and all other points. The projected distances of all point pairs in each view vector direction are calculated using Einstein's summation convention. The formula is as follows:

[0029] ;

[0030] in, It is a point index, with a value range from 0 to N-1; The neighbor index, with values ​​ranging from 0 to N-1, and Not equal to ; To capture the index of the nearest vector, the value ranges from 0 to As-1; Let be the number of three-dimensional coordinates of the point.

[0031] Preferably, in S200, candidate point pairs are screened by applying multiple physical constraints, including:

[0032] Forward projection constraint: Filtering projection distance Point pairs;

[0033] Spatial alignment constraint: Calculate the vertical component distance between pairs of points. ,filter Pairs of points not exceeding a preset alignment threshold; where the vertical component distance... The calculation is as follows:

[0034] ;

[0035] in, Let m be the spherical view vector with index m, where m ranges from 0 to... ;

[0036] Orientation consistency constraint: The orientation consistency metric is obtained by calculating the dot product between the view vectors of the current point and its neighbors. , Filter point pairs whose directional consistency metric value is not less than a preset angle threshold; where, the directional consistency metric value The calculation is as follows:

[0037] ;

[0038] Initial constraint of force closure: Calculate the normal vector of the grabbing viewpoint at the grabbing point. Projection distance on Check the projection relationship between the view vector and the normal vector, and filter. Pairs of points less than or equal to 0; the calculation of the projected distance is as follows:

[0039] ;

[0040] In S200, candidate grasping postures are formed based on the optimal contact point pairs and grasping proximity vectors, including:

[0041] Calculate the translation component using the optimal contact point. This represents the coordinate position of the center point of the grabbing posture in three-dimensional space. The process is as follows:

[0042] ;

[0043] in, and Indicates the coordinate position of the optimal contact point pair;

[0044] The grab proximity vector is converted into a corresponding rotation matrix using the interface functions provided by the graspnetAPI tool library, thus obtaining the rotation representation of the grab pose in the world coordinate system. ;

[0045] Based on translation components and rotational components The grasping posture determines the coordinate position and orientation of the grasp in three-dimensional space.

[0046] Preferably, the calculation of the minimum coefficient of friction in S300 includes:

[0047] S311: Calculate the normal vector for each contact point With the corresponding spherical view vector The actual angle between Specifically:

[0048] ;

[0049] S312: Based on the coefficient of friction Calculate the corresponding friction angle Specifically:

[0050] ;

[0051] S313: For a grasping posture, check whether all its contact points satisfy the condition that the actual included angle is less than or equal to the corresponding friction angle, count the number of effective contact points that satisfy this constraint, and when the number is not less than 3, determine that the grasping posture as a whole satisfies the force closure condition.

[0052] S314: Select all friction coefficients that can satisfy the force closure condition, and take the minimum value among them as the minimum friction coefficient;

[0053] Preferably, the collision detection for grasping posture in S300 specifically includes:

[0054] S321: Calculate the local coordinate position of each point to be processed relative to each grasping posture:

[0055] ;

[0056] in, For each point to be processed, the global coordinates are... For each point to be processed, the local coordinates are... This represents the translation component of the grasping posture, that is, the coordinate position of the grasping posture center point in three-dimensional space. This represents the rotational component of the grasping posture, i.e., the direction of the grasping posture;

[0057] S322: Construct a parameterized simplified gripper model corresponding to the gripping posture. The model includes the left finger, right finger, gripper bottom and proximity area, and generates an axially aligned bounding box based on the geometry of each part.

[0058] S323: By determining the spatial relationship between the local coordinate point cloud and each bounding box, a height mask, left gripper mask, right gripper mask, bottom mask, and moving area mask are generated respectively. The overall collision mask that identifies the collision situation of the point cloud is obtained by combining them through logical operations.

[0059] S324: Calculate the volume of each part of the simplified gripper model, and calculate the Intersection over Union (IoU) value to quantify the severity of the collision using the overall collision mask. Compare the IoU value with the preset collision threshold to generate a binary collision label. If the index exceeds the preset collision threshold, the binary collision label is 1, indicating that there is a collision risk in the gripping posture.

[0060] Preferably, the calculation of the volume of each part of the simplified gripper model in S324 is as follows:

[0061] ;

[0062] ;

[0063] ;

[0064] in, The volume of the left and right fingers. This refers to the volume of the bottom connecting part. The volume of the area where the gripper moves. , , These are the height, length, and width of the grippers. This indicates the safe approach distance that the gripper must maintain before performing a gripping action. Indicates the size of point cloud voxels;

[0065] The total volume of the grippers is:

[0066] ;

[0067] The formula for calculating IoU is:

[0068] ;

[0069] in, It is the intersection-over-union ratio (IoU) metric for collision detection, where N is the total number of points in the point cloud. It is the first The collision mask of a point, if the point A value of 1 indicates a collision with the grabber; otherwise, a value of 0. It is a constant with a value of 1e-6 to prevent division by zero errors.

[0070] Preferably, S400 specifically refers to:

[0071] When the minimum friction coefficient is less than or equal to the set friction coefficient threshold and is positive, the effective gripping width is greater than zero and within the physical limit, and the collision detection result is marked as no collision is possible, then the effective gripping posture is selected.

[0072] A robot optimal grasping contact point matching and grasping quality evaluation system includes:

[0073] The local coordinate system construction module is used to obtain the 3D point cloud of the target object, generate a set of spherical view vectors covering the potential orientation of the object to be grasped based on the 3D point cloud, and generate a grasping proximity vector perpendicular to the spherical view vector.

[0074] The optimal contact point pair matching module is used to calculate the projection distance between each point to be processed on the object surface and its corresponding spherical view vector direction, and all other points in the current spherical view vector direction. By applying multiple physical constraints, candidate point pairs are filtered, and the effective point pair with the largest projection distance is determined as the optimal contact point pair that can form a stable grip under the current view vector direction. The maximum projection distance is recorded, and the corresponding gripping width is determined based on the maximum projection distance. Candidate gripping postures are formed based on the optimal contact point pair and the gripping proximity vector.

[0075] The gripping posture verification module is used to calculate the minimum friction coefficient required to achieve stable gripping in order to evaluate robustness to the smoothness of the object surface, perform collision detection of the gripping posture to ensure operational safety, and verify whether the gripping width is within the physical limits of the gripper.

[0076] The gripping posture determination module is used to comprehensively consider the minimum friction coefficient, collision detection results, and gripping width to select gripping postures that meet the requirements in terms of mechanical stability, collision-free feasibility, and physical feasibility.

[0077] The aforementioned method and system for optimal robot grasping contact point matching and grasping quality assessment systematically constructs a grasping posture search space by predefining a spherical view vector covering the omnidirectional orientation of the object to be grasped, thus improving directional completeness and search efficiency. In the point-pair matching stage, selection is performed based on projection distance and multi-dimensional physical constraints to ensure that the selected contact point pairs possess both maximum grasping width and mechanical stability. In the quality assessment stage, the minimum friction coefficient is comprehensively calculated, collision detection and width verification are conducted, and the feasibility of the grasping posture is comprehensively evaluated from three dimensions: force enclosure, environmental safety, and mechanical constraint. This method can efficiently and reliably generate highly stable, collision-free, and physically executable robot grasping postures. Attached Figure Description

[0078] Figure 1 This is a flowchart of a robot optimal grasping contact point matching and grasping quality evaluation method according to an embodiment of the present invention;

[0079] Figure 2 This is a schematic diagram of a gripper model in one embodiment of the present invention;

[0080] Figure 3This is a flowchart of a contact point matching method in one embodiment of the present invention. Detailed Implementation

[0081] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0082] In one embodiment, such as Figure 1 As shown, a method for optimal robot grasping contact point matching and grasping quality evaluation includes the following steps:

[0083] S100: Obtain the 3D point cloud of the target object, generate a set of spherical view vectors covering the potential orientation of the object to be grasped based on the 3D point cloud, and generate a grasping proximity vector perpendicular to the spherical view vector.

[0084] S200: For each point to be processed on the surface of the object and its corresponding spherical view vector direction, calculate the projection distance of the point to be processed and all other points in the current spherical view vector direction. By applying multiple physical constraints, candidate point pairs are screened. The effective point pair with the largest projection distance is determined as the optimal contact point pair that can form a stable grip under the current view vector direction. The maximum projection distance is recorded. The corresponding gripping width is determined based on the maximum projection distance. Candidate gripping postures are formed based on the optimal contact point pair and the gripping proximity vector.

[0085] S300: Calculates the minimum coefficient of friction required to achieve stable gripping to assess robustness to object surface smoothness, performs collision detection of gripping posture to ensure operational safety, and verifies that the gripping width is within the physical limits of the gripper.

[0086] S400: By combining the minimum coefficient of friction, collision detection results, and gripping width, a gripping posture that meets the requirements in terms of mechanical stability, collision-free feasibility, and physical feasibility is matched.

[0087] The aforementioned method for optimal robot grasping contact point matching and grasping quality evaluation firstly constructs a systematic grasping posture search space by pre-generating a spherical view vector covering the omnidirectional orientation of the object and a grasping proximity vector. This overcomes the limitations of traditional methods, which often suffer from a single or highly random grasping direction, significantly improving the directional completeness and computational efficiency of grasping planning. Secondly, in the contact point matching stage, projection distance is calculated based on the spherical view vector direction, and multi-dimensional physical constraints such as spatial alignment, directional consistency, and force closure are applied for screening. This ensures that the selected contact point pairs not only have the maximum grasping width geometrically but also meet the basic conditions for stable grasping mechanically, improving the success rate and reliability of grasping. Thirdly, in the grasping quality evaluation stage, the minimum friction coefficient is comprehensively calculated, collision detection is performed, and grasping width is verified. Candidate grasping postures are comprehensively evaluated from three dimensions: force closure, environmental safety, and mechanical constraints. This effectively avoids grasping failures caused by smooth surfaces, collisions, or physical impracticality, improving the overall system's robustness and practicality.

[0088] In summary, this method optimizes the entire process from direction generation and point pair matching to comprehensive evaluation. It can efficiently and reliably generate robot grasping postures that are highly stable, collision-free, and physically executable, making it particularly suitable for grasping unknown objects in complex scenarios.

[0089] In one embodiment, a set of uniformly distributed spherical view vectors covering the potential orientation of the object to be grasped is generated on a unit sphere based on the Fibonacci grid sampling algorithm. S100 generating the spherical view vectors includes:

[0090] S110: Create from 0 to The index sequence, calculate the index of each sampling point. The coordinates on the axis are calculated using trigonometric functions. shaft and The coordinate components of the axes are defined by the following formulas:

[0091] ;

[0092] in, This indicates the total number of viewpoints that need to be generated. For indexing, It is an angular parameter related to the golden ratio. , , For the first The coordinate components of each sampling point on the unit sphere;

[0093] S120: Stack the calculated three coordinate components to form a two-dimensional matrix, and map the points on the unit sphere to the actual sphere through scaling and translation transformations, finally obtaining a complete set of spherical view vectors. Specifically:

[0094] ;

[0095] in, Define the three-dimensional coordinates of the sphere's center, which is located at the origin by default. Set to the spherical radius, which defaults to unit length. Further, in this embodiment, the number of generated vectors Vs is 300 spherical viewpoint vectors.

[0096] In one embodiment, generating a grab proximity vector perpendicular to the spherical view vector in S100 includes:

[0097] S130: Ensure that all input normal vectors are unit vectors for each spherical view vector. Using a preset reference vector (To avoid the zero vector problem caused by the normal vector being parallel to the reference vector) an initial grasping proximity vector perpendicular to it is constructed using vector projection. Specifically:

[0098] ;

[0099] in, Represents the multiplication operation between a scalar and a vector;

[0100] S140: When generating the grasp proximity vector, use the Rodriguez rotation formula to make the initial grasp proximity vector... Around the spherical view vector Rotate at preset equal intervals Generate a set of grab proximity vectors that are uniformly distributed in space. The Rodriguez rotation formula is as follows:

[0101] ;

[0102] in, It is the rotation angle. Represents the cross product of vectors. This represents the dot product of vectors. Furthermore, this calculation method ensures that the grasping proximity vectors exhibit a uniform distribution in space. In this embodiment, a batch generation method for vertical vectors is used to generate 12 grasping proximity vectors (As) for each spherical viewpoint vector. The gripper model is as follows: Figure 2 As shown.

[0103] The flowchart of the contact point pair matching method in S200 is as follows: Figure 3 As shown. In one embodiment, S200, calculating the projection distance includes:

[0104] S210: Spherical View Vector Copy and expand to a dimension matching the number N of object gripping points, as the view direction vector originating from each sampling point. This represents the various possible grabbing directions starting from each grabbing point and surrounding that point;

[0105] S220: For each point to be processed, calculate the vector between the current point and all other points. The projected distances of all point pairs in each view vector direction are calculated using Einstein's summation convention. The formula is as follows:

[0106] ;

[0107] in, It is a point index, with a value range from 0 to N-1; The neighbor index, with values ​​ranging from 0 to N-1, and Not equal to ; To capture the index of the nearest vector, the value ranges from 0 to As-1; Let be the number of three-dimensional coordinates of the point.

[0108] In one embodiment, candidate point pairs are screened in S200 by applying multiple physical constraints, including:

[0109] Forward projection constraint: Filtering projection distance The points are aligned to ensure the grabbing direction is reasonable;

[0110] Spatial alignment constraint: Calculate the vertical component distance between pairs of points. ,filter Pairs of points not exceeding a preset alignment threshold; where the vertical component distance... The calculation is as follows:

[0111] ;

[0112] in, Let m be the spherical view vector with index m, where m ranges from 0 to... In addition, the alignment threshold is generally set to 0.005.

[0113] Orientation consistency constraint: The orientation consistency metric is obtained by calculating the dot product between the view vectors of the current point and its neighbors. , Filter point pairs whose directional consistency metric value is not less than a preset angle threshold; where, the directional consistency metric value The calculation is as follows:

[0114] ;

[0115] The directional consistency metric reflects the cosine of the angle between the two view vectors. Furthermore, the angle threshold is typically set to 0.995.

[0116] Initial constraint of force closure: Calculate the normal vector of the grabbing viewpoint at the grabbing point. Projection distance on Check the projection relationship between the view vector and the normal vector, and filter. Pairs of points less than or equal to 0; the calculation of the projected distance is as follows:

[0117] .

[0118] Specifically, if the projected distance is positive, the contact point is in the opposite direction of the normal, the contact force is in the wrong direction, and the force closure condition is not met.

[0119] Furthermore, for each point to be processed and viewpoint, after filtering out valid candidate points that meet the constraints, the point with the largest projected distance is selected as the best matching contact point, forming the optimal contact point pair with the current point. And obtain the maximum projection distance.

[0120] In S200, the corresponding grab width is determined based on the maximum projection distance, including: the initial width is obtained by adding the maximum projection distance to the grab capacity, with the grab capacity set to 0.01 meters by default. Then, the width is limited to an effective range (0.01 to 0.15 meters) by a mask, and values ​​outside the range are set to 0, thus obtaining the effective grab width W.

[0121] In S200, candidate grasping postures are formed based on the optimal contact point pairs and grasping proximity vectors, including:

[0122] Calculate the translation component using the optimal contact point. This represents the coordinate position of the center point of the grabbing posture in three-dimensional space. The process is as follows:

[0123] ;

[0124] in, and Indicates the coordinate position of the optimal contact point pair;

[0125] By using the interface functions provided by the graspnetAPI (an official tool library for robotic grasping research) library, the grasping proximity vector is converted into the corresponding rotation matrix, thereby obtaining the rotational representation of the grasping posture in the world coordinate system. ;

[0126] Based on translation components and rotational components The grasping posture determines the coordinate position and orientation of the grasp in three-dimensional space.

[0127] In one embodiment, calculating the minimum coefficient of friction in S300 includes:

[0128] S311: Calculate the normal vector for each contact point With the corresponding spherical view vector The actual angle between Specifically, the actual angle value is obtained by taking the dot product of the normalized normal vector and the view vector, and then using the inverse cosine function, as shown in the following formula:

[0129] ;

[0130] S312: To evaluate force sealing, based on the relationship between the friction angle and the coefficient of friction, according to the coefficient of friction... Calculate the corresponding friction angle Specifically:

[0131] ;

[0132] The coefficient of friction ranges from 0.1 to 1.

[0133] S313: For a grasping posture, check whether all its contact points satisfy the condition that the actual included angle is less than or equal to the corresponding friction angle, count the number of effective contact points that satisfy this constraint, and when the number is not less than 3, determine that the grasping posture as a whole satisfies the force closure condition.

[0134] S314: Select all friction coefficients that can satisfy the force closure condition, and take the minimum value among them as the minimum friction coefficient.

[0135] Specifically, during the check of force closure conditions, the actual angle between the normal vectors at the contact points is compared with a series of friction angles calculated based on different friction coefficients. The core judgment rule is: for any pair of contact points, the actual angle must be less than or equal to the corresponding friction angle for the pair of contact points to be considered to satisfy the friction constraint of force closure.

[0136] To improve the robustness of the judgment and avoid randomness, the system sets a redundancy threshold: the grasping posture as a whole is judged to satisfy the force closure condition only when the number of effective contact point pairs that satisfy the above constraints in the grasping configuration is not less than 3.

[0137] Finally, all friction coefficients that satisfy the force closure condition are selected, and the minimum value among them is taken. This minimum value is the minimum coefficient of friction required to successfully grasp the target object under the current grasping configuration. It comprehensively reflects the physical feasibility of this set of grasping points under a specific viewing angle.

[0138] In one embodiment, collision detection in the grasping posture in S300 specifically includes:

[0139] S321: Calculate the local coordinate position of each point to be processed relative to each grasping posture:

[0140] ;

[0141] in, For each point to be processed, the global coordinates are... For each point to be processed, the local coordinates are... This represents the translation component of the grasping posture, that is, the coordinate position of the grasping posture center point in three-dimensional space. This represents the rotational component of the grasping posture, i.e., the direction of the grasping posture;

[0142] S322: Construct a parameterized simplified gripper model corresponding to the gripping posture. The model includes the left finger, right finger, gripper bottom and proximity area, and generates an axially aligned bounding box based on the geometry of each part.

[0143] S323: By determining the spatial relationship between the local coordinate point cloud and each bounding box, a height mask, left gripper mask, right gripper mask, bottom mask, and moving area mask are generated respectively. The overall collision mask that identifies the collision situation of the point cloud is obtained by combining them through logical operations.

[0144] Furthermore, collision detection employs a hierarchical bounding box detection strategy. This is achieved by determining the point cloud coordinates. Based on the relative positions of each bounding box, a corresponding binary collision mask is generated to accurately identify the scene points that interfere with the gripper model: the height mask limits the points within the vertical range of the gripper; the left and right gripper masks detect collisions with the gripper fingers, respectively; the bottom mask detects collisions with the bottom of the gripper; and the moving region mask detects potential collisions during gripper movement. These masks are then logically ORed and combined to obtain the overall collision mask. .

[0145] To quantify the severity of collisions, an evaluation metric based on volume occupancy is defined: Intersection over Union (IoU). The point cloud is treated as uniformly distributed sample points in space. The collision volume is approximated by counting the number of points falling into each part of the gripper. The ratio of the collision volume to the total volume of the gripper is then used as the IoU metric.

[0146] S324: Calculate the volume of each part of the simplified gripper model, and calculate the Intersection over Union (IoU) value to quantify the severity of the collision using the overall collision mask. Compare the IoU value with the preset collision threshold to generate a binary collision label. If the index exceeds the preset collision threshold, the binary collision label is 1, indicating that there is a collision risk in the gripping posture.

[0147] In one embodiment, the calculation of the volume of each part of the simplified gripper model in S324 is specifically as follows:

[0148] ;

[0149] ;

[0150] ;

[0151] in, The volume of the left and right fingers. This refers to the volume of the bottom connecting part. The volume of the area where the gripper moves. , , These are the height, length, and width of the grippers. This indicates the safe approach distance that the gripper must maintain before performing a gripping action. Indicates the size of point cloud voxels;

[0152] The total volume of the grippers is:

[0153] ;

[0154] The formula for calculating IoU is:

[0155] ;

[0156] in, It is the intersection-over-union ratio (IoU) metric for collision detection, where N is the total number of points in the point cloud. It is the first The collision mask of a point, if the point A value of 1 indicates a collision with the grabber; otherwise, a value of 0. It is a constant with a value of 1e-6 to prevent division by zero errors.

[0157] In one embodiment, S400 specifically refers to:

[0158] When the minimum friction coefficient is less than or equal to the set friction coefficient threshold and is positive, the effective gripping width is greater than zero and within the physical limit, and the collision detection result is marked as no collision is possible, then the effective gripping posture is selected.

[0159] Specifically, a comprehensive quality assessment is conducted to select valid grabbing candidates. First, the minimum coefficient of friction is considered. A coefficient of friction less than or equal to the threshold means that the grasping posture has sufficient friction to maintain a stable grasp and ensure grasping stability; secondly, it ensures a minimum coefficient of friction. If the value is positive, unstable grabs are excluded; then, the grab width W is kept greater than zero, that is, there is an effective distance between the two contact points; finally, those grab poses that will not collide with other objects in the scene are filtered out, that is, the binary collision label is 0.

[0160] The above-mentioned method for optimal robot grasping contact point matching and grasping quality evaluation has the following beneficial effects:

[0161] (1) By introducing a contact point matching method that comprehensively considers multiple physical constraints such as spatial alignment, directional consistency, and force closure, this invention fundamentally overcomes the problem of inaccurate matching caused by existing technologies that rely solely on a single geometric metric (such as Euclidean distance). This method can match optimal contact point pairs with clear physical meaning and high grasping success rate for robots in complex and occluded scenarios, significantly improving the accuracy of matching and scenario adaptability. It abandons traditional single and one-sided evaluation indicators and constructs a multi-level, multi-factor comprehensive evaluation system for grasping quality. This system, for the first time, quantitatively integrates key physical and practical factors such as the minimum necessary friction coefficient, the rationality of the grasping width, and the risk of scene collisions, achieving a comprehensive and refined evaluation of the stability of the grasping posture. This makes the evaluation results highly consistent with the grasping performance in the real physical world, effectively reducing the risk of grasping failure due to evaluation deviations.

[0162] (2) Due to the introduction of stricter physical and real-world constraints in both the matching and evaluation stages, the candidate grasping posture set generated by this invention has higher average quality and actual success rate. The system can automatically avoid "pseudo-optimal" solutions that are theoretically feasible but physically unstable or prone to collisions, and directly output grasping schemes with high reliability and strong executability. The proposed method does not rely on specific object models or a large amount of labeled data. Its core lies in the encoding of physical principles and the fusion calculation of multi-dimensional constraints. Therefore, it has good generalization ability for unseen object categories and new scenes, reduces the algorithm's dependence on data scale and quality, and is more suitable for complex application scenarios with diverse items and changing environments in actual industries.

[0163] In one embodiment, a robot optimal grasping contact point matching and grasping quality evaluation system is also provided, comprising:

[0164] The local coordinate system construction module is used to obtain the 3D point cloud of the target object, generate a set of spherical view vectors covering the potential orientation of the object to be grasped based on the 3D point cloud, and generate a grasping proximity vector perpendicular to the spherical view vector.

[0165] The optimal contact point pair matching module is used to calculate the projection distance between each point to be processed on the object surface and its corresponding spherical view vector direction, and all other points in the current spherical view vector direction. By applying multiple physical constraints, candidate point pairs are filtered, and the effective point pair with the largest projection distance is determined as the optimal contact point pair that can form a stable grip under the current view vector direction. The maximum projection distance is recorded, and the corresponding gripping width is determined based on the maximum projection distance. Candidate gripping postures are formed based on the optimal contact point pair and the gripping proximity vector.

[0166] The gripping posture verification module is used to calculate the minimum friction coefficient required to achieve stable gripping in order to evaluate robustness to the smoothness of the object surface, perform collision detection of the gripping posture to ensure operational safety, and verify whether the gripping width is within the physical limits of the gripper.

[0167] The gripping posture determination module is used to comprehensively consider the minimum friction coefficient, collision detection results, and gripping width to select gripping postures that meet the requirements in terms of mechanical stability, collision-free feasibility, and physical feasibility.

[0168] Specific limitations regarding the optimal gripping contact point matching and gripping quality assessment system for robots can be found in the limitations of the optimal gripping contact point matching and gripping quality assessment method for robots described above, and will not be repeated here. Each module in the aforementioned optimal gripping contact point matching and gripping quality assessment system for robots can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0169] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a method for optimal robot grasping contact point matching and grasping quality assessment.

[0170] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of a method for optimal robot grasping contact point matching and grasping quality assessment.

[0171] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0172] The above provides a detailed description of the robot optimal grasping contact point matching and grasping quality evaluation method and system provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention, and the descriptions of the embodiments above are only for the purpose of helping to understand the core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A method for optimal robot grasping contact point matching and grasping quality evaluation, characterized in that, The method includes the following steps: S100: Obtain the 3D point cloud of the target object, generate a set of spherical view vectors covering the potential orientation of the object to be grasped based on the 3D point cloud, and generate a grasping proximity vector perpendicular to the spherical view vector. S200: For each point to be processed on the object surface and its corresponding spherical viewpoint vector direction, calculate the projected distance between the point to be processed and all other points in the current spherical viewpoint vector direction. Candidate point pairs are filtered by applying multiple physical constraints. The effective point pair with the largest projected distance is determined as the optimal contact point pair that can form a stable grip under the current viewpoint vector direction, and its maximum projected distance is recorded. The corresponding gripping width is determined based on the maximum projected distance. Candidate gripping postures are formed based on the optimal contact point pair and the gripping proximity vector. In S200, calculating the projected distance includes: S210: Spherical View Vector Copy and expand to a dimension matching the number N of object gripping points, as the view direction vector originating from each sampling point. This represents the various possible grabbing directions starting from each grabbing point and surrounding that point; S220: For each point to be processed, calculate the vector between that point and all other points. The projected distances of all point pairs in each view vector direction are calculated using Einstein's summation convention. The formula is as follows: ； in, It is a point index, with a value range from 0 to N-1; The neighbor index, with values ​​ranging from 0 to N-1, and Not equal to ; To capture the index of the nearest vector, the value ranges from 0 to As-1; The number of three-dimensional coordinates of the point; In S200, candidate point pairs are filtered by applying multiple physical constraints, including: Forward projection constraint: Filtering projection distance Point pairs; Spatial alignment constraint: Calculate the vertical component distance between pairs of points. ,filter Pairs of points not exceeding a preset alignment threshold; where the vertical component distance... The calculation is as follows: ； in, Let m be the spherical view vector with index m, where m ranges from 0 to... ; Orientation consistency constraint: The orientation consistency metric is obtained by calculating the dot product between the view vectors of the current point and its neighbors. , Filter point pairs whose directional consistency metric value is not less than a preset angle threshold; where, the directional consistency metric value The calculation is as follows: ； Initial constraint of force closure: Calculate the normal vector of the grabbing viewpoint at the grabbing point. Projection distance on Check the projection relationship between the view vector and the normal vector, and filter. Pairs of points less than or equal to 0; the calculation of the projected distance is as follows: ； In S200, candidate grasping postures are formed based on the optimal contact point pairs and grasping proximity vectors, including: Calculate the translation component using the optimal contact point. This represents the coordinate position of the center point of the grabbing posture in three-dimensional space. The process is as follows: ； in, and Indicates the coordinate position of the optimal contact point pair; The grab proximity vector is converted into a corresponding rotation matrix using the interface functions provided by the graspnetAPI tool library, thus obtaining the rotation representation of the grab pose in the world coordinate system. ; Based on translation components and rotational components The grasping posture determines the coordinate position and orientation of the grasp in three-dimensional space; S300: Calculates the minimum coefficient of friction required to achieve stable gripping to assess robustness to object surface smoothness, performs collision detection of gripping posture to ensure operational safety, and verifies that the gripping width is within the physical limits of the gripper. S400: Based on the minimum friction coefficient, collision detection results, and gripping width, the gripping posture that meets the requirements in terms of mechanical stability, collision-free feasibility, and physical feasibility is selected.

2. The method according to claim 1, characterized in that, Based on the Fibonacci grid sampling algorithm, a set of uniformly distributed spherical view vectors covering the potential orientation of the object to be grasped is generated on a unit sphere. The spherical view vectors generated by S100 include: S110: Create from 0 to The index sequence, calculate the index of each sampling point. The coordinates on the axis are calculated using trigonometric functions. shaft and The coordinate components of the axes are defined by the following formulas: ； in, This indicates the total number of viewpoints that need to be generated. For indexing, It is an angular parameter related to the golden ratio. , , For the first The coordinate components of each sampling point on the unit sphere; S120: Stack the calculated three coordinate components to form a two-dimensional matrix, and map the points on the unit sphere to the actual sphere through scaling and translation transformations, finally obtaining a complete set of spherical view vectors. Specifically: ； in, Define the three-dimensional coordinates of the sphere's center, which is located at the origin by default. Set to the radius of the sphere; the default value is the unit length.

3. The method according to claim 2, characterized in that, In S100, the grab proximity vector that is perpendicular to the spherical view vector includes: S130: For each spherical view vector Using a preset reference vector An initial grasping proximity vector perpendicular to it is constructed using vector projection. Specifically: ； in, Represents the multiplication operation between a scalar and a vector; S140: When generating the grasp proximity vector, use the Rodriguez rotation formula to make the initial grasp proximity vector... Around the spherical view vector Rotate at preset equal intervals Generate a set of grab proximity vectors that are uniformly distributed in space. The Rodriguez rotation formula is as follows: ； in, It is the rotation angle. Represents the cross product of vectors. Represents the dot product of vectors.

4. The method according to claim 3, characterized in that, The minimum friction coefficient is calculated in S300 including: S311: Calculate the normal vector for each contact point With the corresponding spherical view vector The actual angle between Specifically: ； S312: Based on the coefficient of friction Calculate the corresponding friction angle Specifically: ； S313: For a grasping posture, check whether all its contact points satisfy the condition that the actual included angle is less than or equal to the corresponding friction angle, count the number of effective contact points that satisfy this constraint, and when the number is not less than 3, determine that the grasping posture as a whole satisfies the force closure condition. S314: Select all friction coefficients that can satisfy the force closure condition, and take the minimum value among them as the minimum friction coefficient.

5. The method according to claim 4, characterized in that, The specific aspects of collision detection during grasping posture in S300 include: S321: Calculate the local coordinate position of each point to be processed relative to each grasping posture: ； in, For each point to be processed, the global coordinates are... For each point to be processed, the local coordinates are... This represents the translation component of the grasping posture, that is, the coordinate position of the grasping posture center point in three-dimensional space. This represents the rotational component of the grasping posture, i.e., the direction of the grasping posture; S322: Construct a parameterized simplified gripper model corresponding to the gripping posture. The model includes the left finger, right finger, gripper bottom and proximity area, and generates an axially aligned bounding box based on the geometry of each part. S323: By determining the spatial relationship between the local coordinate point cloud and each bounding box, a height mask, left gripper mask, right gripper mask, bottom mask, and moving area mask are generated respectively. The overall collision mask that identifies the collision situation of the point cloud is obtained by combining them through logical operations. S324: Calculate the volume of each part of the simplified gripper model, and calculate the Intersection over Union (IoU) value to quantify the severity of the collision using the overall collision mask. Compare the IoU value with the preset collision threshold to generate a binary collision label. If the index exceeds the preset collision threshold, the binary collision label is 1, indicating that there is a collision risk in the gripping posture.

6. The method according to claim 5, characterized in that, The specific calculation of the volume of each part of the simplified gripper model in S324 is as follows: ； ； ； in, The volume of the left and right fingers. For the volume of the bottom connecting part, The volume of the area where the gripper moves. , , These are the height, length, and width of the grippers. This indicates the safe approach distance that the gripper must maintain before performing a gripping action. Indicates the size of point cloud voxels; The total volume of the grippers is: ； The formula for calculating IoU is: ； in, It is the intersection-over-union ratio (IoU) metric for collision detection, where N is the total number of points in the point cloud. It is the first The collision mask of a point, if the point A value of 1 indicates a collision with the grabber; otherwise, a value of 0. It is a constant with a value of 1e-6 to prevent division by zero errors.

7. The method according to claim 6, characterized in that, The S400 specifically refers to: When the minimum friction coefficient is less than or equal to the set friction coefficient threshold and is positive, the effective gripping width is greater than zero and within the physical limit, and the collision detection result is marked as no collision is possible, then the effective gripping posture is selected.

8. A robot optimal grasping contact point matching and grasping quality evaluation system, used to perform the method as described in any one of claims 1 to 7, characterized in that, include: The local coordinate system construction module is used to obtain the 3D point cloud of the target object, generate a set of spherical view vectors covering the potential orientation of the object to be grasped based on the 3D point cloud, and generate a grasping proximity vector perpendicular to the spherical view vector. The optimal contact point pair matching module is used to calculate the projection distance between each point to be processed on the object surface and its corresponding spherical view vector direction, and all other points in the current spherical view vector direction. By applying multiple physical constraints, candidate point pairs are filtered, and the effective point pair with the largest projection distance is determined as the optimal contact point pair that can form a stable grip under the current view vector direction. The maximum projection distance is recorded, and the corresponding gripping width is determined based on the maximum projection distance. Candidate gripping postures are formed based on the optimal contact point pair and the gripping proximity vector. The gripping posture verification module is used to calculate the minimum friction coefficient required to achieve stable gripping in order to evaluate robustness to the smoothness of the object surface, perform collision detection of the gripping posture to ensure operational safety, and verify whether the gripping width is within the physical limits of the gripper. The gripping posture determination module is used to comprehensively consider the minimum friction coefficient, collision detection results, and gripping width to select gripping postures that meet the requirements in terms of mechanical stability, collision-free feasibility, and physical feasibility.

Images