A hierarchical clustering-based robot arm grasping stability classification method and system

Abstract

The application provides a kind of based on hierarchical clustering's mechanical arm gripping stability classification method and system, wherein, method includes: step S1: the data obtained by the sensor of mechanical arm is converted into graph data and carries out node embedding;Step S2: inter-cluster clustering is carried out to graph data;Step S3: the interior of each cluster after hierarchical clustering of inter-cluster clustering is layered;Step S4: inter-cluster merging is carried out;Step S5: binary classification is carried out using full connection layer with sigmoid.The based on hierarchical clustering's mechanical arm gripping stability classification method and system of the application can improve the ability of node to obtain deeper feature information, make the clustering result more accurate and the boundary line more clear;Meanwhile, node embedding is carried out using GAT to speed up the speed of automatic learning feature information of node;More accurate and efficient graph structure clustering performance is used in the classification of mechanical arm gripping stability to improve the mechanical stability work rate.

Description

Technical Field

[0001] This invention relates to the field of robotic arm control technology, and in particular to a classification method and system for robotic arm gripping stability based on hierarchical clustering. Background Technology

[0002] With the increasing demand for dexterous grasping and manipulation in industrial and service robots, robot grasping tasks are receiving more and more attention. Regarding the question of whether robotic arms can easily adjust their grasping force in a timely manner based on the weight of an object, reliable prediction of the grasping state can help robotic arms avoid the possibility of grasping failure. Applying depth map clustering algorithms to robotic arms can help them more accurately identify the object's category, thereby effectively improving the speed at which the robotic arm judges the magnitude of force when grasping an object; helping the robot avoid potential grasping failures and having the opportunity to re-grasp the object stably.

[0003] Most graph clustering algorithms are not hierarchical and rely on resolution parameters that allow people to adjust clustering according to the dataset and intended purpose. However, most real-world data consists of complex, multi-scale community structures, and these parameters are difficult to adjust in practice. Furthermore, planar message-passing graph neural networks (GNNs) struggle to capture dependencies between distant node pairs. Inspired by the remarkable effectiveness of deep neural network models in computer vision and natural language processing, researchers later stacked numerous GNN layers together to directly increase the receptive field of each node. Therefore, by simplifying the aggregation design of GNNs and employing carefully designed normalization units or specific gradient descent methods, deeper models were proposed, but the severe over-squeezing problem caused by planar GNNs aggregating information across long paths still exists. Another key drawback of flat message-passing GNNs is that they rely solely on aggregating messages within the observed topology, neglecting information hidden at the meso- and macro-levels. Therefore, this study investigates the idea of ​​a hierarchical message-passing mechanism to enhance the information aggregation pipeline of GNNs. The ultimate goal is to enable the node representation learning process to perceive long-distance interaction information and implicit multi-resolution semantics within the graph. Therefore, hierarchical clustering algorithms are characterized by speed, high memory efficiency, and no parameters. With the introduction and rise of attention mechanisms, which allow GNNs to focus only on the most important or relevant parts of the input data to improve model performance and efficiency, GNNs combined with attention embeddings have also developed rapidly. For example, Graph Attention Networks (GATs) use attention mechanisms to allow nodes to automatically learn the similarity and importance between nodes, rapidly improving the update speed of node weights.

[0004] Many deep learning-based automatic classification methods designed and applied to reliably predict grasping states currently employ a typical approach of feeding tactile representations into a Support Vector Machine (SVM) classifier. However, the current challenges in ensuring the stability of robotic arm grasping lie in the complexity of the data structure and the hidden information between multiple layers of nodes. Furthermore, current techniques suffer from inaccurate clustering results and low efficiency. Summary of the Invention

[0005] One objective of this invention is to provide a hierarchical clustering-based classification method and system for robotic arm gripping stability. This method improves the ability of nodes to acquire deeper feature information while making the clustering results more accurate and the boundaries clearer. Simultaneously, GAT is used for node embedding to accelerate the automatic learning of feature information by the nodes. Applying more accurate and efficient graph structure clustering performance to the classification of robotic arm gripping stability improves the mechanical stability working rate.

[0006] This invention provides a hierarchical clustering-based classification method for classifying the gripping stability of robotic arms, comprising:

[0007] Step S1: Convert the data acquired by the robotic arm's sensors into graph data and embed the nodes;

[0008] Step S2: Perform inter-cluster clustering on the graph data;

[0009] Step S3: Perform hierarchical clustering within each cluster after the inter-cluster clustering is hierarchically divided;

[0010] Step S4: Perform inter-cluster merging;

[0011] Step S5: Perform binary classification using a fully connected layer with a sigmoid function.

[0012] Preferably, step S1: converting the data acquired by the robotic arm's sensors into graph data and embedding nodes, including:

[0013] Step S11: Convert the data acquired by the robotic arm's sensors into graph data, where the graph data is represented as G = (V, E); where V = {V1, V2, ..., V...} N} represents a set of nodes, indicating that a set of graph data has N nodes, and E represents a set of undirected edges;

[0014] Step S12: Apply a shared linear transformation parameterized by the weight matrix W to each node, and perform a self-attention-sharing mechanism on the node, with the attention coefficients as follows:

[0015]

[0016] In the formula, e ijLet be the attention coefficient between the i-th and j-th nodes. These are the feature vectors of the i-th and j-th nodes, respectively;

[0017] Step S13: Normalize all choices at the j-th node using the softmax function.

[0018] Preferably, step S2: performing inter-cluster clustering on the graph data, including:

[0019] Given a weighted undirected graph G = (V, E) with n nodes, let A be the corresponding weighted adjacency matrix; the weight of a node with respect to the i-th and j-th nodes is denoted as the sum of the weights of its associated edges, as shown in the following formula:

[0020] ω i =∑ j∈V A ij ;

[0021] The total weight of the node is:

[0022] ω=∑ i∈V ω i =∑ i,j∈V A ij ;

[0023] The weights of a sampling node to the i-th and j-th nodes can produce a probability distribution:

[0024]

[0025] When considering graph clustering, analogous to the probability distribution of node pairs, the weights induce the probability distribution of cluster pairs:

[0026]

[0027] The distance between two different clusters is expressed as the sampling ratio of the probability; the probability distribution on the cluster and the probability of sampling a given b are:

[0028]

[0029]

[0030] Where p(a|b) is the conditional probability representation of a and b.

[0031] Preferably, step S3: performing hierarchical clustering within each cluster after inter-cluster clustering stratification, including:

[0032] Treating each cluster as a super node, a supergraph is designed at layer t, with the following hyperparameter γ:

[0033]

[0034] Intra-cluster hierarchical clustering consists of three steps at level t: bottom-up propagation, intermediate layer propagation, and top-down propagation. The updated representation of the supernodes at level t during bottom-up propagation is as follows:

[0035]

[0036] in, It is a t-layer supergraph G t The supernode in s t-1 It belongs to layer t-1 The node, Is it Figure G? t Zhongyou Layer generation Node feature representation, Indicates belonging to a supernode The number of nodes in layer t-1;

[0037] The expression for aggregating information from layer t through intermediate layer propagation is as follows:

[0038]

[0039] In the formula, Is node u in The feature representation after bottom-up propagation of layers, N t (v) is the set of nodes adjacent to node v at level t. It is an aggregated node representation of v based on local neighborhood information;

[0040] For continuously updated node representations, an attention mechanism is used to adaptively learn the contribution weights of different levels during the top-down propagation process.

[0041]

[0042] In the formula α uv MEAN is the trainable normalized attention coefficient between node v and super node u or between node v and itself, and C(v) represents the set of super nodes from different layers of node v from {2,…,k}.

[0043] The output node of the last layer is generated as follows:

[0044]

[0045] For each node in each layer, a bottom-up propagation, intermediate layer propagation, and top-down propagation are performed sequentially to obtain the representation vector of each node. The model is then trained using a loss function based on cross-entropy; the loss function is as follows:

[0046]

[0047] Preferably, the hierarchical clustering-based classification method for robotic arm gripping stability further includes:

[0048] Before performing inter-cluster merging, acquire the robot arm's parameter data and usage environment data as analysis data;

[0049] Based on the analysis data and the pre-set analysis database, a benchmark for inter-cluster merging is determined;

[0050] Among these, the benchmarks for inter-cluster merging are determined based on the analyzed data and a pre-set analysis database, including:

[0051] Feature extraction is performed on the analyzed data, and a benchmark analysis set is constructed based on the extracted feature values;

[0052] Match the benchmark analysis set with the standard set associated with each benchmark in the analysis database;

[0053] Based on the matching results, the corresponding benchmark is retrieved.

[0054] This invention also provides a robotic arm gripping stability classification system based on hierarchical clustering, comprising:

[0055] The data conversion and embedding module is used to convert the data acquired by the robotic arm's sensors into graph data and embed nodes.

[0056] The inter-cluster clustering module is used to perform inter-cluster clustering on graph data;

[0057] The hierarchical clustering module is used to perform hierarchical clustering within each cluster after the inter-cluster clustering is hierarchically stratified;

[0058] The merge module is used for inter-cluster merging;

[0059] The classification module is used for binary classification using a fully connected layer with a sigmoid function.

[0060] Preferably, the data conversion and embedding module converts the data acquired by the robotic arm's sensors into graph data and embeds nodes, performing the following operations:

[0061] The data acquired by the robotic arm's sensors is converted into graph data, which is represented as G = (V, E); where V = {V1, V2, ..., V...} N} represents a set of nodes, indicating that a set of graph data has N nodes, and E represents a set of undirected edges;

[0062] A shared linear transformation parameterized by the weight matrix W is applied to each node, and a self-attention-sharing mechanism is performed on the nodes, with the attention coefficients as follows:

[0063]

[0064] In the formula, e ij Let be the attention coefficient between the i-th and j-th nodes. These are the feature vectors of the i-th and j-th nodes, respectively;

[0065] Normalization is performed on all choices at the j-th node using the softmax function.

[0066] Preferably, the inter-cluster clustering module performs inter-cluster clustering on the graph data, executing the following operations:

[0067] Given a weighted undirected graph G = (V, E) with n nodes, let A be the corresponding weighted adjacency matrix; the weight of a node with respect to the i-th and j-th nodes is denoted as the sum of the weights of its associated edges, as shown in the following formula:

[0068] ω i =∑ j∈V A ij ;

[0069] The total weight of the node is:

[0070] ω=∑ i∈V ω i =∑ i,j∈V A ij ;

[0071] The weights of a sampling node to the i-th and j-th nodes can produce a probability distribution:

[0072]

[0073] When considering graph clustering, analogous to the probability distribution of node pairs, the weights induce the probability distribution of cluster pairs:

[0074]

[0075] The distance between two different clusters is expressed as the sampling ratio of the probability; the probability distribution on the cluster and the probability of sampling a given b are:

[0076]

[0077] Where p(a|b) is the conditional probability representation of a and b.

[0078] Preferably, the hierarchical clustering module performs hierarchical clustering within each cluster after the inter-cluster clustering is hierarchically stratified, and performs the following operations:

[0079] Treating each cluster as a super node, a supergraph is designed at layer t, with the following hyperparameter γ:

[0080]

[0081] Intra-cluster hierarchical clustering consists of three steps at level t: bottom-up propagation, intermediate layer propagation, and top-down propagation. The updated representation of the supernodes at level t during bottom-up propagation is as follows:

[0082]

[0083] in, It is a t-layer supergraph G t The supernode in s t-1 It belongs to layer t-1 The node, Is it Figure G? t Zhongyou Layer generation Node feature representation, Indicates belonging to a supernode The number of nodes in layer t-1;

[0084] The expression for aggregating information from layer t through intermediate layer propagation is as follows:

[0085]

[0086] In the formula, Is node u in The feature representation after bottom-up propagation of layers, N t (v) is the set of nodes adjacent to node v at level t. It is an aggregated node representation of v based on local neighborhood information;

[0087] For continuously updated node representations, an attention mechanism is used to adaptively learn the contribution weights of different levels during the top-down propagation process.

[0088]

[0089] In the formula α uv MEAN is the trainable normalized attention coefficient between node v and super node u or between node v and itself, and C(v) represents the set of super nodes from different layers of node v from {2,…,k}.

[0090] The output node of the last layer is generated as follows:

[0091]

[0092] For each node in each layer, a bottom-up propagation, intermediate layer propagation, and top-down propagation are performed sequentially to obtain the representation vector of each node. The model is then trained using a loss function based on cross-entropy; the loss function is as follows:

[0093]

[0094] Preferably, the robotic arm gripping stability classification system based on hierarchical clustering further includes: a benchmark determination module;

[0095] The benchmark determination module performs the following operations:

[0096] Before performing inter-cluster merging, acquire the robot arm's parameter data and usage environment data as analysis data;

[0097] Based on the analysis data and the pre-set analysis database, a benchmark for inter-cluster merging is determined;

[0098] Among these, the benchmarks for inter-cluster merging are determined based on the analyzed data and a pre-set analysis database, including:

[0099] Feature extraction is performed on the analyzed data, and a benchmark analysis set is constructed based on the extracted feature values;

[0100] Match the benchmark analysis set with the standard set associated with each benchmark in the analysis database;

[0101] Based on the matching results, the corresponding benchmark is retrieved.

[0102] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0103] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0104] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0105] Figure 1 This is a schematic diagram of a robotic arm gripping stability classification method based on hierarchical clustering in an embodiment of the present invention;

[0106] Figure 2 This is a schematic diagram of the object-grabbing steps of the robotic arm used in this invention;

[0107] Figure 3 This is a simplified schematic diagram of the classification steps of the present invention;

[0108] Figure 4 This is a schematic diagram of a robotic arm gripping stability classification system based on hierarchical clustering in an embodiment of the present invention. Detailed Implementation

[0109] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0110] This invention provides a hierarchical clustering-based classification method for classifying the gripping stability of robotic arms, such as... Figure 1 As shown, it includes:

[0111] Step S1: Convert the data acquired by the robotic arm's sensors into graph data and embed the nodes;

[0112] Step S2: Perform inter-cluster clustering on the graph data;

[0113] Step S3: Perform hierarchical clustering within each cluster after the inter-cluster clustering is hierarchically divided;

[0114] Step S4: Perform inter-cluster merging;

[0115] Step S5: Perform binary classification using a fully connected layer with a sigmoid function.

[0116] The working principle and beneficial effects of the above technical solution are as follows:

[0117] Figure 2 This is a flowchart illustrating the object-retrieving process of the robotic arm used in the classification method of this invention. Figure 3 This is a flowchart of the hierarchical clustering algorithm of the present invention; the data acquired by the robotic arm sensors is converted into graph data, and the graph is represented as a tuple G = (V, E), where V = {V1, V2, ..., V...} N Let} represent a set of nodes, indicating a group with N nodes, and E represent a set of undirected edges. First, GAT is used for node embedding to dynamically capture the correlations between node features. The formula used is as follows:

[0118]

[0119] The initial step is to apply a shared linear transformation parameterized by the weight matrix W to each node, and then perform a self-attention-sharing mechanism α,e on the nodes. ij Let be the attention coefficient between nodes i and j, where Let i be the feature vectors of nodes i and j.

[0120]

[0121] To make these attention coefficients easy to compare across different nodes, the softmax function is used to normalize them across all choices of node j.

[0122]

[0123] Since the attention mechanism α is a single-layer feedforward neural network, therefore α ij These are real coefficients calculated for the fully expanded attention mechanism. By stacking multiple graph attention layers, a complete GAT network can be obtained for node embedding.

[0124] The complexity and redundancy of the node data are reduced by embedding it using GAT. The next step is to apply a hierarchical clustering algorithm based on probability distribution to the embedded data. This algorithm first constructs a hierarchical structure for the embedded data using the idea of ​​probability distribution. Then, it performs hierarchical clustering on each cluster. The supergraph at level t undergoes a three-step propagation process: bottom-up propagation, intermediate layer propagation, and top-down propagation. Finally, it merges clusters, thus achieving more accurate and efficient clustering of the graph data. The specific steps are as follows:

[0125] First, we use the idea of ​​probability distribution to perform inter-cluster clustering on a set of graph data. Given a weighted undirected graph G = (V, E) with n nodes, let A be the corresponding weighted adjacency matrix, which is a symmetric non-negative matrix. For a pair of nodes i and j, the weight of node i is denoted as the sum of the weights of its associated edges, as shown in the following formula:

[0126]

[0127] The total weight of the node is:

[0128]

[0129] Using ideas from probability theory, the weights of a sampling node with respect to i and j can generate a probability distribution:

[0130]

[0131] Similarly, when considering the clustering of a graph C (i.e., a partition of V), analogous to the probability distribution of node pairs, the weights induce the probability distribution of cluster pairs:

[0132]

[0133] Representing the distance between two different clusters using the sampling ratio of probabilities allows for more precise numerical calculation of the inter-cluster distance, thereby improving clustering performance. The probability distribution on the clusters and the probability of sampling a given b are as follows:

[0134]

[0135]

[0136] Where p(a|b) represents the conditional probability of a and b. The distance calculated by the above formula will be used in the algorithm to merge the closest clusters.

[0137] Modular clustering is a graph-based clustering method that aims to divide a graph into multiple subgraphs with high connectivity density within each subgraph and low connectivity density between subgraphs. This method can be used to discover community structures in a graph, i.e., sets of nodes that are similar or interdependent. Modularity is a standard metric for evaluating the quality of cluster C (any partition of V). Modular clustering techniques generally rely on the definition of modularity, but the main drawback of this formula is its resolution limitation. To overcome this limitation, a multiplication factor γ, called resolution, is introduced. Therefore, modularity is:

[0138]

[0139] Introducing a probability distribution representation, the modularity of cluster C with a multiplication factor γ is:

[0140]

[0141] As the modularity Qγ(C) increases, C is updated by moving a node from one cluster to another. The target cluster for each node is to maximize modularity. This algorithm can be viewed as a modularity maximization scheme with sliding resolution. The desired modularity maximization scheme can then be quantified: starting with the maximum resolution of each node's own cluster, find the first value of the resolution parameter γ, such as γ1. The resolution at time t (the time when the t-th merge is triggered) is 1 / dt, corresponding to clustering C. t Since the resolution sequence does not increase, the algorithm provides a complete hierarchical structure.

[0142]

[0143] To achieve better clustering results, after inter-cluster clustering, hierarchical clustering is performed within each of the stratified clusters. The main method is to treat each cluster as a super node and design a super graph at level t. The hyperparameter γ is calculated using equation (12). Intra-cluster hierarchical clustering consists of three steps at level t: bottom-up propagation, intermediate layer propagation, and top-down propagation. The expressions for each propagation method are as follows:

[0144]

[0145] Equation (13) is the updated representation of the supernode at layer t during bottom-up propagation, where It is a t-layer supergraph G t The supernode in s t-1

[0146] It belongs to layer t-1 The node, Is it Figure G? t Zhongyou Layer generation Node feature representation, Indicates belonging to a supernode The number of nodes in layer t-1.

[0147]

[0148] Equation (14) is the expression for the aggregation of information at layer t propagated from the intermediate layer. Wherein, Is node u in The feature representation after bottom-up propagation of layers, N t (v) is the set of nodes adjacent to node v at level t. It is an aggregated node representation of v based on local neighborhood information.

[0149] For continuously updated node representations, an attention mechanism is used to adaptively learn the contribution weights of different levels during the top-down propagation process.

[0150]

[0151] In equation (15) α uv Let V be the trainable normalized attention coefficient between node v and supernode u, or between v and itself. MEAN is calculated as the element-wise mean. C(v) represents the set of supernodes from different layers of node v from {2,…,k}. The output node generation of the last layer (L) is represented by Equation (16):

[0152]

[0153] Loss function determination: For each node in each layer, perform bottom-up propagation, intermediate layer propagation, and top-down propagation sequentially to obtain the representation vector of each node. Then, train the model using a loss function based on cross-entropy. The loss function is as follows:

[0154]

[0155] This paper applies a hierarchical clustering algorithm based on probability distribution to the dataset acquired by the robotic arm's sensors. At the end of the model, a fully connected layer (FC) with a sigmoid function is used for binary classification. Various sensors control the robotic arm's gripping force to achieve the gripping task. By reliably predicting the gripping state, the robotic arm can effectively avoid potential gripping failures and has the opportunity to reliably grasp objects. Furthermore, applying depth map clustering algorithms to the robotic arm helps it more accurately identify the object's category, thereby effectively improving the speed at which the robotic arm determines the force required to grasp objects.

[0156] In one embodiment, before performing inter-cluster merging, parameter data of the robotic arm and usage environment data are obtained as analysis data;

[0157] Based on the analysis data and the pre-set analysis database, a benchmark for inter-cluster merging is determined;

[0158] Among these, the benchmarks for inter-cluster merging are determined based on the analyzed data and a pre-set analysis database, including:

[0159] Feature extraction is performed on the analyzed data, and a benchmark analysis set is constructed based on the extracted feature values;

[0160] Match the benchmark analysis set with the standard set associated with each benchmark in the analysis database;

[0161] Based on the matching results, the corresponding benchmark is retrieved.

[0162] The working principle and beneficial effects of the above technical solution are as follows:

[0163] The determination of the baseline is fundamental to the merging process and determines its accuracy. The robotic arm's parameter data includes: model, movement radius, maximum gripping force, number of grippers, and relative positional relationships. Environmental data includes: temperature, humidity, electromagnetic interference data, and sound intensity data. The feature extraction step involves extracting each data point and quantifying it to form corresponding quantified values ​​as feature values. These feature values ​​are then arranged in a predetermined order to form the baseline analysis set. The analysis database is pre-configured by professionals, with a one-to-one correspondence between the standard set and the baselines. By matching the baseline analysis set with the standard set, the baseline can be retrieved, adapting to different environments and determining the merging baseline for the robotic arms. For example, the baseline includes a threshold and a judgment condition, such as merging those less than or equal to the threshold. The matching between the standard set and the baseline analysis set can be achieved by calculating their similarity, specifically using the cosine similarity calculation method.

[0164] To further improve the accuracy of the robotic arm in judging the magnitude of force when grasping an object, a pressure detection array can be used to detect the gravity distribution on two adjacent sides of the object.

[0165] Construct a gravity distribution matrix based on the gravity distribution.

[0166] Based on the gravity distribution matrix and classification results, the grasping force is determined from the preset grasping library;

[0167] Among them, based on the gravity distribution, a gravity distribution matrix is ​​constructed, including:

[0168] The positioning line is based on the edge where the two measured sides of the object meet. The sensor data of the gravity distribution detected on the two sides are stitched together. Then, the X-axis is determined with any direction of the positioning line as the X-axis direction, and the Y-axis is determined perpendicular to the X-axis direction. The origin is the midpoint of the positioning line.

[0169] Determine the coordinate positions corresponding to each detection data point;

[0170] Based on the coordinate positions of the detection data corresponding to each side, and based on the preset feature point determination rules, the feature points are determined.

[0171] Extract pressure data from feature points and fill them into a preset gravity distribution matrix template to form a gravity distribution matrix.

[0172] The rules for determining feature points are configured in advance by professionals; for example: the point corresponding to the coordinate position with the shortest sum of the coordinate positions corresponding to all other detection data is taken as the midpoint; the point with the farthest distance from the midpoint is taken as the reference point from the four perpendicular directions; the point with the smallest sum of distances from the reference point and the midpoint is taken as the intermediate point; the midpoint, intermediate point and reference point are taken as feature points.

[0173] Specifically, based on the gravity distribution matrix and classification results, the grasping force is determined from a pre-set grasping library, including:

[0174] Based on the classification results, the corresponding crawling intensity sub-library is determined from the preset crawling library;

[0175] The gravity distribution matrix is ​​matched with each standard matrix in the gripping force sub-library to obtain the corresponding gripping force determination result.

[0176] Each grasping force determination sub-library corresponds to one type or class of objects. The classification result indicates that the object has been determined, and the corresponding grasping force determination sub-library can be directly called. The grasping force determination results in the grasping force determination sub-library are associated with the standard matrix. The grasping force determination results can be retrieved by directly matching the standard matrix with the gravity distribution matrix. The matching can be achieved by calculating the similarity between the two matrices.

[0177] This embodiment uses a pressure detection array to detect the gravity distribution on two adjacent sides of an object and then analyzes and determines the gravity distribution of the object, thereby improving the accuracy of the robotic arm in judging the magnitude of force when grasping an object.

[0178] This invention also provides a classification system for robotic arm gripping stability based on hierarchical clustering, such as... Figure 4 As shown, it includes:

[0179] Data conversion and embedding module 1 is used to convert the data acquired by the robotic arm's sensors into graph data and embed nodes;

[0180] Inter-cluster clustering module 2 is used to perform inter-cluster clustering on graph data;

[0181] Hierarchical clustering module 3 is used to perform hierarchical clustering within each cluster after inter-cluster clustering is hierarchically stratified;

[0182] Merging module 4 is used for inter-cluster merging;

[0183] Classification module 5 is used for binary classification using a fully connected layer with a sigmoid function.

[0184] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A hierarchical clustering-based grasp stability classification method for a robot arm, characterized by, include: Step S1: Convert the data acquired by the robotic arm's sensors into graph data and embed the nodes; Step S2: Perform inter-cluster clustering on the graph data; Step S3: Perform hierarchical clustering within each cluster after the inter-cluster clustering is hierarchically divided; Step S4: Perform inter-cluster merging; Step S5: Perform binary classification using a fully connected layer with a sigmoid function; Step S1: Converting the data acquired by the robotic arm's sensors into graph data and embedding nodes, including: Step S11: converting data acquired by the sensor of the mechanical arm into graph data, wherein the graph data is represented as ; in the formula, is a node set, indicating that a group of graph data has nodes, indicates a group of undirected edges; Step S12: Apply a shared linear transformation parameterized by the weight matrix W to each node, and perform a self-attention-sharing mechanism on the node, with the attention coefficients as follows: ； In the formula, is an attention coefficient between the i-th and j-th nodes, is a feature vector of the i-th and j-th nodes, respectively; Step S13: Normalize all choices at the j-th node using the softmax function.

2. The hierarchical clustering-based robotic arm gripping stability classification method as described in claim 1, characterized in that, Step S2: Perform inter-cluster clustering on the graph data, including: Given a weighted undirected graph with n nodes, let A be the corresponding weighted adjacency matrix; for a pair of nodes, the ith and jth, the weight is the sum of the weights of the edges associated with them, as follows: ； The total weight of the node is: ； The weights of a sampling node to the i-th and j-th nodes can produce a probability distribution: ； When considering graph clustering, analogous to the probability distribution of node pairs, the weights induce the probability distribution of cluster pairs: ； The distance between two different clusters is expressed as the sampling ratio of the probability; the probability distribution on the cluster and the probability of sampling a given b are: = ； ； wherein is the conditional probability of a, b.

3. The hierarchical clustering based robotic arm grasp stability classification method of claim 2, wherein, Step S3: Perform hierarchical clustering within each cluster after inter-cluster clustering stratification, including: Each cluster is regarded as a super node, a super graph is designed at t layer, and the super parameter at this time is As follows: ； Intra-cluster hierarchical clustering consists of three steps at level t: bottom-up propagation, intermediate layer propagation, and top-down propagation. The updated representation of the supernodes at level t during bottom-up propagation is as follows: ； in, It is a t-layer supergraph G t super nodes in It belongs to layer t-1 The node, Is it Figure G? t Zhongyou -1 layer generated Node feature representation, Indicates belonging to a supernode The number of nodes in layer t-1; The expression for aggregating information from layer t through intermediate layer propagation is as follows: ； In the formula, Is node u in Feature representation after bottom-up propagation of layer -1 Let t be the set of nodes adjacent to node v. It is an aggregated node representation of v based on local neighborhood information; For continuously updated node representations, an attention mechanism is used to adaptively learn the contribution weights of different levels during the top-down propagation process. ； In the formula Let V be the trainable normalized attention coefficient between node v and supernode u, or between v itself. MEAN is calculated as the element-wise mean. Indicates from {2, The set of supernodes at different levels of node v in the k}; The output node of the last layer is generated as follows: ； For each node in each layer, a bottom-up propagation, intermediate layer propagation, and top-down propagation are performed sequentially to obtain the representation vector of each node. The model is then trained using a loss function based on cross-entropy; the loss function is as follows: 。 4. The hierarchical clustering based robotic arm grasp stability classification method of claim 3, wherein, Also includes: Before performing inter-cluster merging, acquire the robot arm's parameter data and usage environment data as analysis data; Based on the analysis data and the preset analysis database, a benchmark for inter-cluster merging is determined; The criteria for inter-cluster merging are determined based on the analyzed data and a preset analysis database, including: Feature extraction is performed on the analyzed data, and a benchmark analysis set is constructed based on the extracted feature values; Match the benchmark analysis set with the standard set associated with each benchmark in the analysis database; Based on the matching results, the corresponding benchmark is retrieved.

5. A hierarchical clustering based robotic arm grasp stability classification system, characterized by, include: The data conversion and embedding module is used to convert the data acquired by the robotic arm's sensors into graph data and embed nodes. The inter-cluster clustering module is used to perform inter-cluster clustering on graph data; The hierarchical clustering module is used to perform hierarchical clustering within each cluster after the inter-cluster clustering is hierarchically stratified; The merge module is used for inter-cluster merging; The classification module is used for binary classification using a fully connected layer with a sigmoid function. The data conversion and embedding module converts the data acquired by the robotic arm's sensors into graph data and embeds nodes, performing the following operations: The data acquired by the robotic arm's sensors is converted into graph data, where the graph data is represented as follows: In the formula, A node set represents a set of graph data. 1 node Represents a set of undirected edges; A shared linear transformation parameterized by the weight matrix W is applied to each node, and a self-attention-sharing mechanism is performed on the nodes, with the attention coefficients as follows: ； In the formula, is an attention coefficient between the i-th and j-th nodes, is a feature vector of the i-th and j-th nodes, respectively. Normalization is performed on all choices at the j-th node using the softmax function.

6. The hierarchical clustering based robotic arm grasp stability classification system of claim 5, wherein, The inter-cluster clustering module performs inter-cluster clustering on graph data, executing the following operations: Given a weighted undirected graph with n nodes, let A be the corresponding weighted adjacency matrix; for a pair of nodes, the ith and jth, the weight is the sum of the weights of the edges associated with them, as follows: ； The total weight of the node is: ； The weights of a sampling node to the i-th and j-th nodes can produce a probability distribution: ； When considering graph clustering, analogous to the probability distribution of node pairs, the weights induce the probability distribution of cluster pairs: ； The distance between two different clusters is expressed as the sampling ratio of the probability; the probability distribution on the cluster and the probability of sampling a given b are: = ； ； wherein is the conditional probability of a, b.

7. The hierarchical clustering based robotic arm grasp stability classification system of claim 6, wherein, The hierarchical clustering module performs hierarchical clustering within each cluster after the inter-cluster clustering is hierarchically stratified, and performs the following operations: Treating each cluster as a supernode, a supergraph is designed at layer t. The hyperparameters at this point are... as follows: ； Intra-cluster hierarchical clustering consists of three steps at level t: bottom-up propagation, intermediate layer propagation, and top-down propagation. The updated representation of the supernodes at level t during bottom-up propagation is as follows: ； in, It is a t-layer supergraph G t super nodes in It belongs to layer t-1 The node, Is it Figure G? t Zhongyou -1 layer generated Node feature representation, Indicates belonging to a supernode The number of nodes in layer t-1; The expression for aggregating information from layer t through intermediate layer propagation is as follows: ； In the formula, Is node u in Feature representation after bottom-up propagation of layer -1 Let t be the set of nodes adjacent to node v. It is an aggregated node representation of v based on local neighborhood information; For continuously updated node representations, an attention mechanism is used to adaptively learn the contribution weights of different levels during the top-down propagation process. ； wherein is a trainable normalized attention coefficient between node v and supernode u or itself, MEAN is an element-wise mean computation, denotes a set of different layer supernodes of node v from {2, ,k}. The output node of the last layer is generated as follows: ； For each node in each layer, a bottom-up propagation, intermediate layer propagation, and top-down propagation are performed sequentially to obtain the representation vector of each node. The model is then trained using a loss function based on cross-entropy; the loss function is as follows: 。 8. The hierarchical clustering based robotic arm grasp stability classification system of claim 7, wherein, Also includes: Benchmark determination module; The benchmark determination module performs the following operations: Before performing inter-cluster merging, acquire the robot arm's parameter data and usage environment data as analysis data; Based on the analysis data and the preset analysis database, a benchmark for inter-cluster merging is determined; The criteria for inter-cluster merging are determined based on the analyzed data and a preset analysis database, including: Feature extraction is performed on the analyzed data, and a benchmark analysis set is constructed based on the extracted feature values; Match the benchmark analysis set with the standard set associated with each benchmark in the analysis database; Based on the matching results, the corresponding benchmark is retrieved.

Images