A robot path planning method for fiber flexible body operation

Abstract

The application discloses a robot path planning method for fiber flexible body operation. In view of the low efficiency of the traditional manual fiber winding method, the application provides a robot path planning method for fiber flexible body operation, and the method can improve the production efficiency of the optical fiber sensor and reduce the manufacturing cost by adopting the robot to replace the manual fiber winding. Through actual verification, the method disclosed by the application can accurately control the optical fiber winding according to the designed path, realizes the intelligent fiber winding, and has high application value and economic benefits.

Description

TECHNICAL FIELD

[0001] The present application relates to a flexible body shape control and path planning method in the optical fiber industry, and belongs to the field of flexible body shape control and path planning. BACKGROUND

[0002] Optical fiber sensor is the most important and widely used sensor among all sensors. In the production process of optical fiber sensor, the optical fiber connection task includes optical fiber stripping, optical fiber cleaning, optical fiber cutting, and optical fiber fusion, among which the optical fiber coiling is the most time-consuming and labor-intensive link.

[0003] At present, the coiling link in the optical fiber connection task still adopts the traditional manual coiling method, and the coiling efficiency cannot meet the industrial demand, which seriously hinders the market development of optical fiber sensors. With the expansion of the application field of robots and the change of actual demand, robots can also complete the control operation of flexible deformable objects at the present stage. It is an extremely effective method to use robots to replace manual coiling, and its performance fully meets the actual requirements of industrial coiling. SUMMARY

[0004] The purpose of the present application is to provide a method for controlling optical fiber coiling by robot to improve the production efficiency of optical fiber sensor and reduce the manufacturing cost.

[0005] In order to achieve the above purpose, the technical scheme of the present application provides a robot path planning method for fiber flexible body operation, characterized in that it comprises the following steps:

[0006] Step 1: randomly move the optical fiber end held by the end gripper of the mechanical arm, obtain the point cloud data of the optical fiber image collected by the three-dimensional camera and the pose of the end gripper of the mechanical arm;

[0007] Step 2: input the real-time obtained point cloud data of the optical fiber image into the trained mechanism and data fusion network model, output the speed information of each key point of the optical fiber at the current t-1 moment from the mechanism and data fusion network model Further based on obtain the position information X of each key point of the optical fiber at the next moment t t , wherein X t is the position information of each key point of the optical fiber at the current t-1 moment, and Δt is the step length of one time step;

[0008] Step 3: set the pre-coiling position close to the coiling box as the target position X d , repeat step 2 to repeatedly solve the optimal robot motion using model predictive control, and send the result to the robot for execution until ||X(t)-X d||≤e, where e is a preset target threshold, so as to move the randomly placed optical fiber to the position of the pre-fiber disc;

[0009] Step 4: Collect point cloud data of the fiber disc box with a three-dimensional camera, and use color filtering to screen out target buckle point clouds. By calculating the centroid of each screened area, a representative position of each buckle is obtained. Meanwhile, the principal component analysis method is used to find the main axis by analyzing the variability of the point cloud data to determine the direction of different target buckles. By comprehensively considering the position and direction of the buckles, the order of the optical fiber passing through the target buckles is determined to generate an initial reference trajectory;

[0010] Step 5: Construct a virtual magnetic vector field ring at each target buckle position. By dynamically adjusting the gripper position and recalculating the rope motion path, each buckle is inserted in turn according to the initial reference trajectory, which includes the following steps:

[0011] Step 501: Control the gripper to make the optical fiber tip pass through the first target buckle. Set the virtual magnetic vector field to ensure that all integral curves pass through the buckle opening. The direction of the vector field is designed so that all field lines can enter the opening from one side and leave from the other side, ensuring that there is a closed solution form at any position in the workspace, thereby speeding up the calculation process. The magnetic field induced by the current-carrying ring constructed in accordance with these characteristics is shown below:

[0012]

[0013] In the formula, p represents any point in the workspace, S represents a fixed current-carrying ring, and F(p) represents the magnetic force at point p. The position g of the gripper grabbing the optical fiber is calculated using the relationship between the opening direction of the first target ring and the direction of gravity. The relationship between the motion of the gripper encoded by the network model and the key points, as well as the magnetic field equation of the target ring, are used to calculate the speed of the optical fiber tip to predict the expected motion of the key points on the optical fiber

[0014] Step 502: If the difference between the expected motion and the actual motion exceeds the tolerance value, adjust the gripper to be close to the key point to ensure that the gripper moves according to the expected path.

[0015] Step 503: After the gripper pushes the optical fiber through the first target buckle, it needs to be re-grabbed. When calculating the grabbing position, the angle θ between the normal of the target ring and the direction of gravity is calculated to determine the distance l / θ between the gripper grip point and the final key point. After the optical fiber tip penetrates the target virtual ring, the optical fiber is selected to be re-grabbed or not. This re-grabbing method is combined with the gripper, and the optical fiber is passed through a series of planned buckles in this way to achieve intelligent fiber discing.

[0016] ​Preferably, in step 2, the mechanism and data fusion network model is built based on a graph neural network model, and a mechanism module is embedded in the graph neural network model.

[0017] Preferably, in step 2, training the mechanism and data fusion network model comprises the following steps:

[0018] Step 201: The obtained point cloud data of the optical fiber image is regarded as a sample from a Gaussian mixture model, and the centroid of the Gaussian mixture model is the key point position X(t). According to Bayes' theorem, a sampling point is drawn from the Gaussian mixture model The probability is expressed as:

[0019]

[0020] In the formula, p(n) represents the weight of the nth mixed component, represents the corresponding probability of sampling from the mixed component;

[0021] The problem is converted into a maximum likelihood estimation problem, in which the mixed centroid is optimized so as to maximize the logarithmic likelihood L of the point cloud data Y t sampling, that is:

[0022]

[0023] In the formula: sigma represents the standard deviation of the covariance matrix of each component in the Gaussian mixture model; represents the centroid position of the nth mixed component at time t; S represents the total number of sample points in the point cloud data; p(n) represents the weight of the nth mixed component; N+1 represents the index of the component in the mixed model, wherein N is the number of Gaussian components of the real data points, and N+1 is an additional component for modeling noise and outliers;

[0024] By the Gaussian mixture model and Bayesian inference, the positions of the key points of the collected point cloud data are calculated, and then the obtained pose data of the end gripper of the mechanical arm are corresponded, so as to obtain a training image data set;

[0025] Step 202: The relationship between the key points and the end gripper pose is obtained by constructing an optical fiber mechanism model through abquas software, and the relationship data between the key points and the end gripper pose are used to initialize the weights of the mechanism and data fusion network. The Neo-Hookean model is used to represent the simplest form of strain potential energy in the optical fiber mechanism model, and the energy function of the Neo-Hookean model is as follows:

[0026]

[0027] where U is the strain energy per unit reference volume; C 10 , D1 is the material parameter of the optical fiber; is the first deviatoric strain invariant, J el is the elastic volume ratio;

[0028] Step 203: training the mechanism and data fusion network model by the training image dataset, during the training, first passing through the encoding module, the encoder processes the information of the vertices and edges of the graph G, projects the graph G to a low-dimensional representation G=(V, E) to improve efficiency, wherein the vertices V correspond to the position sequence of the key points of the optical fiber, the edges E correspond to the interaction between the key points, the output of the encoder is the low-dimensional representation of the graph, i.e. the encoded vertex feature 0 v i ′ and edge feature 0 e i ′ ,j , which are taken as the input of the processor module;

[0029] The processor module is composed of multiple information transmission modules, the first information transmission module receives the output from the encoder, i.e. the low-dimensional vertex feature 0 v i ′ and edge feature 0 e i ′ ,j ; the interaction between the vertices in the latent graph is calculated to simulate the dynamics of the optical fiber, and the output after passing through the last message passing module is the updated latent vertex feature K+1 v i ′ and edge feature K+ 1 e i ′ ,j , which are taken as the input of the decoder; finally, the obtained latent representation is restored to the prediction of the original optical fiber motion through the decoder module, and the output is the velocity information of each key point of the optical fiber at a specific time where, based on the current graph state and the control input of the mechanical arm, the process of propagating the graph vertex and edge features in the processor module is represented as follows:

[0030]

[0031] k+1 e i ′ ,j =f e ( k e i ′ ,j , k v i ′, k v j ′)

[0032] where: fv (·) and f e (·) represent graph vertex network and graph edge network respectively; k v i ′, k e i ′ ,j represent current vertex and edge features; k+1 v i ′, k+1 e i ′ ,j represent latent vertex and edge features after k-th message passing; represents the sum of features of edges connected to vertex i; k v i ′ and k v′ j are features of two vertices i and j connected by edge (i, j) respectively.

[0033] Preferably, the step 4 comprises the following steps:

[0034] Step 401: Perform a decentralization operation on the data set D composed of point cloud data, as shown in the following formula:

[0035]

[0036] In the formula, m is the number of data, X = [x1, x2, …, xn], each x is an n-dimensional column vector. m

[0037] Step 402: Calculate the covariance matrix C of the data set D, as shown in the following formula:

[0038]

[0039] Find a direction in the above data space such that the variance of the projection of the data in this direction is maximum, where w is the direction with maximum projection variance, and w T x i is the coordinate value of each data in this direction. is the covariance matrix of the sample, denoted as C, then the optimization objective is as shown in the following formula:

[0040]

[0041] Step 403: Perform eigenvalue decomposition on the covariance matrix C, and the obtained eigenvector indicates the main direction of the data, and the eigenvalue represents the variation degree along each direction. Use the Lagrange constant method to solve, then:

[0042] L(w, λ) = w T Cw + λ (1-w T w)​

[0043]

[0044] Solving for:

[0045]

[0046] maxD(x)=max{w T Cw}=max{w T λw}=maxλ

[0047] In the formula: w is the eigenvector of C;

[0048] Find the largest variance, which is also the largest eigenvalue of the covariance matrix. The direction at this point is the eigenvector corresponding to the largest eigenvalue. Select the first few eigenvectors with the largest eigenvalue. These eigenvectors point to the main direction of data change, which is also the main direction of the fixed device.

[0049] To address the inefficiency of traditional manual fiber coiling methods, this invention proposes a robotic path planning method for flexible fiber optic applications. By employing robots to replace manual fiber coiling, the production efficiency of fiber optic sensors can be improved and manufacturing costs reduced. Practical verification demonstrates that the method disclosed in this invention can precisely control fiber coiling according to a pre-designed path, achieving intelligent fiber coiling and possessing high application value and economic benefits. Attached Figure Description

[0050] Figure 1 This is a flowchart of the algorithm of the present invention;

[0051] Figure 2 This illustrates the data and mechanism fusion network model in the fiber optic path planning method of the present invention;

[0052] Figure 3 This is a virtual magnetic vector field construction diagram in the fiber optic path planning method of the present invention. Detailed Implementation

[0053] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0054] like Figure 1 As shown in the figure, an embodiment of the present invention discloses a robot path planning method for fiber flexible body operations, which specifically includes the following steps:

[0055] Step 1: Randomly move the fiber end held by the end gripper of the robot arm, obtain the point cloud data of the fiber image captured by the three-dimensional camera and the pose of the end gripper of the robot arm.

[0056] Step 2: Input the real-time obtained point cloud data of the fiber image into the trained mechanism and data fusion network model, and output the shape prediction of the fiber by the mechanism and data fusion network model.

[0057] In combination Figure 2 , the mechanism and data fusion network model is built based on a graph neural network model (GNN), and a mechanism module is embedded in the backbone network of the GNN. Training the mechanism and data fusion network model includes the following steps:

[0058] Step 201: The obtained point cloud data is regarded as samples from a Gaussian Mixture Model (GMM), and the centroid is the key point position X(t). According to Bayes' theorem, the probability of sampling a point from the Gaussian mixture model can be expressed as:

[0059]

[0060] In the formula, p(n) represents the weight of the nth mixture component, represents the corresponding probability of sampling from the mixture component.

[0061] Then the problem is converted into a maximum likelihood estimation (MLE) problem, where we want to optimize the mixture centroid to maximize the log-likelihood L of the sampled point cloud data Y t , then we have:

[0062]

[0063] In the formula: σ represents the standard deviation of the covariance matrix of each component in the Gaussian mixture model; represents the centroid position of the nth mixture component (or Gaussian component) at time t, which is the key point position we want to optimize by maximum likelihood estimation; S represents the total number of sample points in the point cloud data, that is, the number of all samples considered when calculating the probability; p(n) represents the weight of the nth mixture component; N+1 represents the index of the component in the mixture model, where N is the number of Gaussian components of the real data points, and N+1 is an additional component for modeling noise and outliers.

[0064] By Gaussian mixture model and Bayesian inference, the position of the key point of the collected point cloud data is calculated, and the obtained pose data of the end gripper of the robot arm is corresponded, thereby obtaining the training image data set.

[0065] Step 202: Construct the optical fiber mechanism model by abquas software to obtain the relationship between the key points and the end gripper pose, and initialize the mechanism and data fusion network weight by referencing the obtained relationship data between the key points and the end gripper pose. The optical fiber mechanism model adopts a Neo-Hookean model to represent the simplest form of strain potential energy, and the energy function of the Neo-Hookean model is as follows:

[0066]

[0067] In the formula, wherein, U is the strain energy of the unit reference volume; C 10 , D1 is the material parameter of the optical fiber; is the first deviatoric strain invariant, J el is the elastic volume ratio.

[0068] Step 203: Train the mechanism and data fusion network model by the training image data set. During training, the information of the vertices and edges of the graph G is processed by the encoding module. The graph G is projected to a low-dimensional representation G=(V,E) to improve efficiency, wherein the vertices V correspond to the position sequence of the optical fiber key points, and the edges E correspond to the interaction between the key points. The output of the encoder is the low-dimensional representation of the graph, i.e. the encoded vertex features 0 v i ′ and edge features 0 e i ′ ,j . These outputs will be used as the input of the first information transmission module.

[0069] Then, the first information transmission module receives the output from the encoder, i.e. the low-dimensional vertex features 0 v i ′ and edge features 0 e i ′ ,j , through the processor module (composed of multiple information transmission modules). The interaction between the vertices in the latent graph is calculated to simulate the dynamics of the optical fiber, and the output after the last message passing module is the updated latent vertex features K+1 v i ′ and edge features K+1 e i ′ ,j , which will be used as the input of the decoder. Finally, the obtained latent representation is restored to the prediction of the original optical fiber motion through the decoder module, and the output is the velocity information of each key point of the optical fiber at a specific time , which is the target output of the entire model. Thus, the training of the mechanism and data fusion network model is completed. Among them, based on the current graph state and the control input of the mechanical arm, the process of propagating the graph vertex and edge features in the processor module is represented as follows:

[0070]

[0071] k+1 e i ′ ,j =f e ( k e i ′ ,j , k v i ′, k v j ′)

[0072] In the formula: f v (·) and f e (·) represent the vertex network and edge network of the graph, respectively; k v i ′, k e i ′ ,j Indicates the current vertex and edge features; k+1 v i ′, k+1 e i ′ ,j Represent the potential vertex and edge features after the k-th message passing; This represents the sum of the characteristics of the edges connected to vertex i; k v i 'and k v′ j These are the characteristics of the two vertices i and j connected by the edge (i,j).

[0073] Step 204: Substitute the obtained velocities at each key point into the formula. To obtain the key point location of the fiber optic cable at the next time step, we set the pre-coil position near the fiber optic cable cassette as our target position X. d Model predictive control (MPC) is used to iteratively solve for the optimal robot motion, and the results are sent to the robot for execution until ||X(t)-X d ||≤e. This allows for the movement of randomly placed optical fibers to the pre-coiled fiber position.

[0074] Step 3: Acquire point cloud data of the fiber optic cable box using a 3D camera, and use color filtering to select the target clip point cloud. Calculate the centroid of each selected region as the representative position of each clip. Simultaneously, use Principal Component Analysis (PCA) to analyze the variability of the point cloud data to find the principal axis and determine the orientation of different target clips. By comprehensively considering the position and orientation of the clips, the order in which the optical fiber passes through the target clips is determined, generating an initial reference trajectory. The algorithm steps are as follows:

[0075] Step 301: Perform a decentering operation on the data set D, as shown in the following equation:

[0076]

[0077] where m is the number of data, X = [x1, x2,..., xn], each xi is an n-dimensional column vector; m

[0078] Step 302: Calculate the covariance matrix C of the data set D, as shown in the following equation:

[0079]

[0080] Our goal is to find a direction in this data space such that the variance of the projection of the data in this direction is maximized. In this equation, w is the direction with the maximum projection variance, w T x i is the coordinate value of each data in this direction. C is the covariance matrix of the sample. Our optimization goal is:

[0081]

[0082] Step 303: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvectors, which indicate the main directions of the data, and the eigenvalues, which represent the variation along each direction. We use the Lagrange multiplier method to solve it.

[0083] L(w, λ) = w T Cw + λ(1 - w T w)

[0084]

[0085] The solution is:

[0086]

[0087] maxD(x) = max{w T Cw} = max{w T λw} = max λ

[0088] In this equation, w is the eigenvector of C. Finding the maximum variance is equivalent to finding the maximum eigenvalue of the covariance matrix, and the direction at this time is the eigenvector corresponding to the maximum eigenvalue. Select the first few eigenvectors with the largest eigenvalues, which point to the main variation directions of the data, i.e., the main directions of the fixture.

[0089] ​Step 4: Construct a virtual magnetic vector field ring at each target buckle position, and sequentially insert each buckle according to the initial reference trajectory by dynamically adjusting the gripper position and recalculating the rope motion path. Step 4 specifically includes the following steps:

[0090] Step 401: Control the gripper to pass the first target buckle with the fiber tip. We set up a virtual magnetic vector field to ensure that all integral curves pass through the buckle opening, so that the fiber can be easily inserted into the buckle opening regardless of the initial configuration. The direction of the vector field is designed so that all field lines can enter the opening from one side and exit from the other side, ensuring that there is a closed solution form at any position in the workspace, thereby speeding up the calculation process. The magnetic field induced by the current-carrying ring we construct that follows these characteristics is shown below:

[0091]

[0092] where p represents any point in the workspace, S represents the fixed current-carrying ring, and F(p) represents the magnetic force at point p. Using the relationship between the opening direction of the first target ring and the direction of gravity, we calculate the position g at which the gripper grasps the fiber. By using the relationship between the gripper motion encoded by the network model and the key point motion, as well as the magnetic field equation of the target ring, we calculate the fiber tip speed to predict the desired motion of the key points on the fiber

[0093] Step 402: If the difference between the expected motion and the actual motion exceeds the tolerance value, the gripper needs to be adjusted to be close to the key points to ensure that the gripper moves according to the expected path.

[0094] Step 403: Because the gripper cannot pass through the buckle, it needs to be re-grasped after the gripper pushes the fiber through the first target buckle. When calculating the grasping position, we determine it by calculating the angle θ between the target ring normal and the direction of gravity. The gripper grasping point is selected to be l / θ (l is a constant that can be determined experimentally) away from the final key point. After the fiber tip penetrates the target virtual ring, we can choose to re-grasp or not to grasp the fiber, depending on the situation. This re-grasping method is combined with the gripper to pass the fiber through a series of planned buckles, achieving intelligent fiber winding.

Claims

1. A robot path planning method for operations on flexible fiber bodies, characterized in that, Includes the following steps: Step 1: The end of the optical fiber is held by the end gripper of the robotic arm and moved randomly to obtain point cloud data of the optical fiber image captured by the 3D camera and the pose of the end gripper of the robotic arm. Step 2: Input the point cloud data of the real-time acquired fiber optic image into the trained mechanism and data fusion network model, and the mechanism and data fusion network model will output the velocity information of each key point of the fiber optic cable at the current time t-1. Furthermore based on Obtain the position information X of each key point of the optical fiber at the next time t. t , where X t This represents the position information of each key point in the optical fiber at the current time t-1, where Δt is the step size of one time step; Step 3: Set the pre-coiling position near the fiber optic cassette as the target position X. d Repeat step 2, using model predictive control to repeatedly solve for the optimal robot motion, and send the results to the robot for execution, until ||X(t)-X d ||≤e, thus enabling the arbitrary placement of optical fibers to be moved to the pre-coiled position, where e is the preset target threshold; Step 4: Use a 3D camera to acquire point cloud data of the fiber optic cable box, and use color filtering to select the target buckle point cloud. Calculate the centroid of each selected region as the representative position of each buckle. At the same time, use principal component analysis to find the main axis by analyzing the variability of the point cloud data to determine the direction of different target buckles. By comprehensively considering the position and direction of the buckles, determine the order in which the optical fiber passes through the target buckles to generate the initial reference trajectory. Step 5: Construct a virtual magnetic vector field loop at each target buckle position. By dynamically adjusting the gripper position and recalculating the rope motion path, insert each buckle sequentially according to the initial reference trajectory. This includes the following steps: Step 501: Control the clamp to guide the fiber tip through the first target latch; set up a virtual magnetic vector field to ensure that all integral curves pass through the latch opening. The direction of the vector field is designed so that all field lines can enter the opening from one side and exit from the other side, ensuring a closed solution form at any position in the workspace, thereby accelerating the calculation process. The magnetic field induced by the current-carrying ring, which follows these characteristics, is constructed as follows: In the formula, p represents any point in the workspace, S represents a fixed current-carrying ring, and F(p) represents the magnetic force at point p. The position g of the gripper holding the optical fiber is calculated using the relationship between the opening direction of the first target ring and the direction of gravity. The relationship between the gripper and the motion of the key points, encoded by the network model, and the magnetic field equation of the target ring are used to calculate the velocity at the end of the optical fiber to predict the desired motion of the key points on the optical fiber. Step 502: If there is a difference between the desired motion and the actual motion If the tolerance value is exceeded, the gripper is readjusted to move closer to the critical point to ensure that the gripper moves along the expected path; Step 503: After the holder pushes the optical fiber past the first target clip, it needs to be gripped again. When calculating the gripping position, it is determined by calculating the angle θ between the target ring normal and the direction of gravity. The gripping point of the holder is selected to be beyond the final key point by a distance l / θ. After the optical fiber tip penetrates the target virtual ring, the holder can choose to grip or not grip the optical fiber. This re-gripping method is combined with the holder. The optical fiber is passed through a series of planned clips in this way to achieve intelligent fiber coiling.

2. The robot path planning method for flexible fiber body operations as described in claim 1, characterized in that, In step 2, the mechanism and data fusion network model is built based on a graph neural network model, and a mechanism module is embedded in the graph neural network model.

3. The robot path planning method for flexible fiber body operations as described in claim 2, characterized in that, Step 2, training the mechanism and data fusion network model includes the following steps: Step 201: Treat the point cloud data of the obtained fiber optic image as samples from a Gaussian mixture model, with the centroid at the keypoint location X(t). According to Bayes' theorem, extract sampling points from the Gaussian mixture model. probability Represented as: In the formula, p(n) represents the weight of the nth mixture component. Indicates sampling from the mixed components The corresponding probability; The problem is transformed into a maximum likelihood estimation problem, where the hybrid centroid is optimized to maximize the point cloud data Y. t Given the sampled log-likelihood L, we have: In the formula: σ represents the standard deviation of the covariance matrix of each component in the Gaussian mixture model; denoted by t, where represents the centroid position of the nth mixture component at time t; S represents the total number of sample points in the point cloud data; p(n) represents the weight of the nth mixture component; N+1 represents the index of the component in the mixture model, where N is the number of Gaussian components in the real data points, and N+1 is an additional component used to model noise and outliers. By using Gaussian mixture model and Bayesian inference, the positions of key points in the collected point cloud data are calculated, and then matched with the pose data of the robotic arm end effector to obtain the training image dataset. Step 202: Construct a fiber optic mechanism model using Abquas software to obtain the relationship between keypoints and end effector pose. Use this relationship data to initialize the weights of the mechanism and data fusion network. This fiber optic mechanism model uses the Neo-Hookean model to represent the simplest form of strain potential energy. The energy function of this Neo-Hookean model is as follows: In the formula, U is the strain energy per unit reference volume; C 10 D1 is the material parameter of the optical fiber; It is the first partial strain invariant, J el It is the elastic volume ratio; Step 203: Train the mechanism and data fusion network model using the training image dataset. During training, the model first passes through an encoding module. The encoder processes the vertex and edge information of graph G, projecting graph G onto a low-dimensional representation G = (V, E) to improve efficiency. Here, vertex V corresponds to the position sequence of fiber optic keypoints, and edge E corresponds to the interaction between keypoints. The encoder output is the low-dimensional representation of the graph, i.e., the encoded vertex features. 0 v′ i Sum of edge features 0 e′ i,j These outputs serve as inputs to the processor module; The processor module consists of multiple information transmission modules. The first information transmission module receives the output from the encoder, namely the low-dimensional vertex features. 0 v′ i Sum of edge features 0 e′ i,j The dynamics of an optical fiber are simulated by calculating the interactions between vertices in the latent graph. The output after passing through the last message-passing module is the updated latent vertex features. K+1 v′ i Sum of edge features K+1 e′ i,j The latent representation is used as input to the decoder; finally, the decoder module restores the obtained latent representation to the original fiber motion prediction, and the output is the velocity information of each key point of the fiber at a specific moment. The process of propagating graph vertex and edge features in the processor module, based on the current graph state and the control input of the robotic arm, is expressed as follows: k+1 e′ i,j =f e ( k e′ i,j , k v′ i , k v′ j ) In the formula: f v (·) and f e (·) represent the vertex network and edge network of the graph, respectively; k v′ i , k e′ i,j Indicates the current vertex and edge features; k+1 v′ i , k+1 e′ i,j Represent the potential vertex and edge features after the k-th message passing; This represents the sum of the characteristics of the edges connected to vertex i; k v′ i and k v′ j These are the characteristics of the two vertices i and j connected by the edge (i,j).

4. The robot path planning method for flexible fiber body operations as described in claim 1, characterized in that, Step 4 includes the following steps: Step 401: Perform a decentralization operation on the dataset D composed of point cloud data, as shown in the following equation: In the formula: m is the number of data points, X = [x1, x2, ..., x m Each x is an n-dimensional column vector; Step 402: Calculate the covariance matrix C of dataset D, as shown in the following formula: In the dataset D above, find a direction that maximizes the variance of the projection of the data along that direction, where w is the direction with the maximum projection variance. T x i The coordinate value of each data point in this direction; Let C be the covariance matrix of the sample. Then the optimization objective is as follows: Step 403: Perform eigenvalue decomposition on the covariance matrix C. The resulting eigenvectors indicate the main directions of the data, while the eigenvalues ​​represent the degree of variation along each direction. Solving using the Lagrange daily number method, we have: L(w,λ)=w T Cw+λ(1-w T In) Solving for: maxD(x)=max{w T Cw}=max{w T λw}=maxλ In the formula: w is the eigenvector of C; Find the largest variance, which is also the largest eigenvalue of the covariance matrix. The direction at this point is the eigenvector corresponding to the largest eigenvalue. Select the first few eigenvectors with the largest eigenvalue. These eigenvectors point to the main direction of data change, which is also the main direction of the fixed device.

Images