Intelligent clamping jaw structure for mounting overhead line system cantilever and control method

Abstract

The invention discloses an intelligent clamping jaw structure for mounting a cantilever of a contact network and a control method, and the method comprises the steps: carrying out the registration matching of real-time scanned point cloud data and a preset cantilever three-dimensional model through an improved iterative nearest point algorithm, carrying out the smooth processing of a registration result through the dynamic adjustment of a data frame in an optimization window, and carrying out the smooth processing of the data frame. According to the method, the jump problem possibly occurring in the single-frame registration process in the high-speed movement of the cantilever is effectively inhibited, the data of the inertial measurement unit is innovatively and directly brought into the construction process of the loss function, and the registration precision and robustness are further improved. By optimizing the size of the window, the balance between the real-time performance and the precision requirement is realized, and the space coordinates and the attitude angle of the insulator are accurately extracted. Visual data are mapped to a mechanical arm control coordinate system through a coordinate transformation matrix, and real-time data synchronization of a visual system and a clamping jaw adjusting device is ensured. The device is suitable for various types of cantilevers, the automation degree is high, the precision is high, and the mounting efficiency is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent assembly of catenary cantilevers in rail transit, and in particular to an intelligent gripper structure and control method for installing catenary cantilevers. Background Art

[0002] With the rapid development of technology, the installation technology of catenary cantilevers in rail transit is also constantly improving. Many new technologies, such as machine vision, laser sensors, intelligent learning and other technologies, have been widely applied to the installation of cantilevers, making the automated installation technology of catenary cantilevers continuously advancing and evolving.

[0003] In the prior art, the installation of catenary cantilevers mainly relies on manual labor. The general technological process is as follows: First, hang a pulley at the top of the column, fix the pre-assembled cantilever with a rope, the ground personnel pull the cantilever to the top of the column, and the personnel above insert the cantilever into the base and manually insert bolts and tighten nuts. A total of 6 - 7 people are required, and the installation time is about 10 minutes, which consumes a large amount of physical strength of the construction personnel. Especially in harsh environments such as plateaus and deserts, physical exhaustion is serious. Therefore, it is very necessary to carry out research on the automation of the cantilever installation process and create an intelligent installation process for catenary cantilevers.

[0004] Although mechanization has also appeared in the current installation process, there are mainly problems in the construction equipment level, such as the low coverage rate of the entire mechanized process. Especially when grasping and installing the cantilever, the grippers in the prior art cannot be applied to multiple models of cantilevers, and cannot automatically insert bolts and tighten nuts. Summary of the Invention

[0005] In order to overcome the above problems existing in the prior art, the present invention proposes an intelligent gripper structure and control method for installing catenary cantilevers.

[0006] The technical solution adopted by the present invention to solve its technical problems is: A control method for an intelligent gripper structure for installing catenary cantilevers, including a dynamic calibration stage and a closed-loop control stage. In the dynamic calibration stage, the vision system periodically scans the target position, detects the offset of the cantilever during the installation process, and triggers a dynamic adjustment instruction for the gripper posture. In the closed-loop control stage, an optimized control algorithm based on dynamic tuning of fuzzy PID parameters is adopted. Combining the position and angle deviation amounts fed back by the vision system, the output torque of the cylinder and the offset of the guide rail are adjusted in real time; An improved iterative closest point algorithm is used to register and match the point cloud data scanned in real time with the preset three-dimensional model of the cantilever, dynamically adjust the data frames within the optimization window, smooth the registration results, and directly incorporate the data of the inertial measurement unit into the construction process of the loss function. Set a fault tolerance mechanism during the control process. When the vision system fails due to occlusion, strong light interference, or abnormal data, switch to the safe mode, predict the target position by interpolating historical motion trajectory data, and activate redundant sensors for auxiliary positioning.

[0007] The above-mentioned intelligent gripper structure control method for catenary boom installation, the specific real-time adjustment of the cylinder output torque and the guide rail displacement is as follows: Step 1, take the friction coefficient of the clamping surface , the real-time inclination angle of the boom and the clamping force deviation e as input variables; Step 2, PID parameter dynamic adjustment and anti-saturation strategy: Introduce a momentum factor , weight and fuse the historical adjustment amount and the current error change, and calibrate the value through experiments to balance real-time response and parameter stability; ; Among them, is the value of the proportional factor at time t + 1, is the change amount of the proportional factor, and t is the current time step.

[0008] Introduce an anti-saturation strategy. When the clamping force approaches the boom compression threshold, the integral term is frozen: ; Among them, is the control gain, is the boom compression threshold, is the actual value of the clamping force; Step 3, construct an energy function , where , ensure that its derivative satisfies by designing the control law: ; Among them, is the derivative of the energy function, is the error change rate.

[0009] The above-mentioned intelligent gripper structure control method for catenary boom installation, the specific quantization method of the input variables in Step 1 is as follows: Friction coefficient : Obtained in real time through a spectral sensor, divided into low, medium, and high grades, where the low grade , the medium grade , the high grade , and its membership function adopts a Gaussian distribution: ; Among them, is the membership function of the low grade; is the membership function for the medium level; is the membership function for the high level; The inclination angle of the boom : calculated by the vision system, and the membership function is trapezoidally distributed: ; The clamping force deviation e: the target clamping force and the actual value The difference, that is ; Its fuzzy set is {negative large, negative small, zero, positive small, positive large}, and the membership function adopts triangular distribution.

[0010] The above intelligent gripper structure control method for catenary boom installation, the improved iterative closest point algorithm is specifically as follows: through the rotation matrix R and the translation vector s, set the initial spatio-temporal loss weights and , the initial transformation matrix , define the sliding window size as N, indicating the pose parameters of the joint optimization of the nearest N frames; During iterative calculation, for each point , search for the bidirectional nearest neighbor point in the reference point cloud Q, and introduce the dynamic distance threshold : ; Among them, , is the initial threshold, is the attenuation coefficient, and k is the number of iterations; To suppress the influence of outliers, introduce the weighted centroid based on the matching confidence and , is the noise standard deviation; Construct the covariance matrix jointly through multiple frames of data within the sliding window, perform weighted SVD decomposition on the covariance matrix , and update the rotation matrix and the translation vector; Define the spatio-temporal loss function , fuse the geometric error and the motion consistency, and realize the spatio-temporal joint optimization of the sliding window; Dynamically adjust the spatio-temporal loss weights and ; Solve the minimization problem of the spatio-temporal loss function and update the transformation matrix of all frames within the window; Determine whether to converge based on the mean square error change, the maximum number of iterations, and the stability of the transformation matrix. If it converges, output the optimized current frame transformation matrix. If it does not converge, use the IMU data to predict the initial pose of the next frame.

[0011] The above intelligent gripper structure control method for catenary boom installation, the weighted centroid and The calculation formula is: ; where i is the three-dimensional coordinate of the i-th point, from the source point cloud p, N is the total number of coordinates, is the weight coefficient of the i-th point in the k-th iteration, and j is the three-dimensional coordinate of the j-th point, from the target point cloud q.

[0012] The above intelligent gripper structure control method for catenary boom installation, the covariance matrix The calculation formula is: ; The spatio-temporal loss function The calculation formula is: ; ; where, is deduced from the IMU data, is the number of points participating in the matching at time t, is the current motion speed, is the adjustment coefficient.

[0013] An intelligent gripper structure for catenary boom installation, used to implement the above intelligent gripper structure control method for catenary boom installation, including a vision sensor, a back plate, a clamping structure, and a bolt and nut picking structure. The vision sensor is installed on the back plate, and at least three groups of clamping structures are installed on the back plate. The clamping structures are arranged in a triangle. One of the clamping structures is installed on the back plate through an adjustable slide rail. The bolt and nut picking structure is installed on the side of the clamping structure corresponding to the end of the catenary boom where bolts and nuts need to be installed.

[0014] The above intelligent gripper structure for catenary boom installation, the bolt and nut picking structure includes a mounting frame, a bolt picking structure, and a nut picking structure. The bolt picking structure is installed on the mounting frame through a sliding structure. The nut picking structure corresponds to the bolt picking structure to ensure the matching of bolts and nuts. The bolt picking structure and the nut picking structure both realize picking through electromagnetic force.

[0015] The above-mentioned intelligent gripper structure for catenary boom installation, the adjustable slide rail includes a linear slide rail and an arc slide rail. The linear slide rail is installed on the back plate, and the arc slide rail is installed on the linear slide rail. The arc slide rail can slide up and down along the linear slide rail. The clamping structure is installed on the arc slide rail and can slide along the arc slide rail.

[0016] The beneficial effect of the present invention is that the improved iterative closest point algorithm (ICP algorithm) of the present invention is used to register and match the point cloud data scanned in real time with the preset three-dimensional model of the boom. The data frames in the optimization window are dynamically adjusted to smooth the registration result, effectively suppressing the jump problem that may occur in the single-frame registration process during the high-speed movement of the boom. And innovatively incorporate the inertial measurement unit (IMU) data directly into the construction process of the loss function to further improve the accuracy and robustness of the registration. In addition, the size of the sliding window can be flexibly adjusted according to the specific application scenario. By optimizing the window size, the balance between real-time performance and accuracy requirements is achieved, providing strong support for the precise control of the boom in the high-speed movement scenario, so as to accurately extract the spatial coordinates (X, Y, Z) and attitude angle θ of the insulator. The visual data is mapped to the manipulator control coordinate system through the coordinate transformation matrix to ensure the real-time data synchronization between the visual system and the gripper adjustment device, providing precise guidance for subsequent actions.

[0017] The intelligent gripper structure of the present invention is applicable to various models of booms, realizing the grasping of the boom, intelligently identifying the column base, automatically inserting it into the base, and automatically inserting the fastening nut, with high automation and high precision, effectively improving the installation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of the intelligent gripper structure of the present invention; Figure 2 is a schematic diagram of the bolt and nut picking structure of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0019] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be described in detail below with reference to the drawings and specific embodiments.

[0020] As Figure 1As shown in the figure, this embodiment discloses an intelligent jaw structure for the installation of catenary boom, which includes a backplane 1, a clamping structure 2, and a bolt and nut picking structure 4. At least three groups of clamping structures 2 (three groups are installed in this embodiment) are installed on the backplane 1. The three groups of clamping structures are arranged in a triangle. One of the clamping structures is installed on the backplane 1 through an adjustable slide rail 3. The bolt and nut picking structure 4 is installed on the side of the clamping structure corresponding to the end of the catenary boom where bolts and nuts need to be installed. The clamping structure is provided with torque by a cylinder to clamp the flat boom and the inclined boom. To maintain stability, once the boom is clamped, it will remain motionless. To ensure safety, during the clamping operation, in case of faults such as power failure or air supply interruption, the clamping structure is equipped with a safety locking device and can still lock the boom tightly to ensure that it will not fall off. The adjustable slide rail includes an arc slide rail and a linear slide rail. The linear slide rail is installed on the backplane, and the arc slide rail is installed on the linear slide rail. The arc slide rail can slide up and down along the linear slide rail. The clamping structure is installed on the arc slide rail and can slide along the arc slide rail. The arc slide rail can meet the variation adjustment within the range of 25° - 40° of the variation angle, and the linear slide rail can meet the up and down adjustment, so as to meet the variation within the range of 1.6 meters - 1.85 meters of the upper and lower spacing of the column base. The backplane is a metal plate for installing each component. To meet the requirements of safety and lightness during operation, the backplane in this embodiment is processed from aviation aluminum and adopts a frame structure.

[0021] As Figure 2 shown, the bolt and nut picking structure includes a mounting bracket 6, a bolt picking structure 8, and a nut picking structure 11. The bolt picking structure 8 is installed on the mounting bracket through a sliding structure 7. The nut picking structure corresponds to the bolt picking structure to ensure the matching of bolts and nuts. The bolt picking structure and the nut picking structure both achieve picking through electromagnetic force. The nut picking structure 11 has a function of adjustable torque to complete the nut tightening under the set torque condition. In this embodiment, the adjustable torque function can be achieved by an electric wrench (or servo motor) with a settable torque value. After setting the torque value, the electric wrench (or servo motor) receives a signal trigger and tightens the nut under the set torque.

[0022] The bolt picking structure 8 and the nut picking structure 11 pick up the bolt 9 and the nut 10 by electromagnetic force. After the clamping structure grabs the wrist arm and inserts it into the base, the bolt and nut picking structure starts to operate. There is a fixed distance between the bolt and nut picking structure and the screw holes of the upper and lower bases. At this time, the sliding structure 7 drives the bolt picking structure to move, inserts the bolt into the screw hole, and then the nut picking structure 11 fastens the nut to the bolt. At this time, the nut is fastened according to the preset torque. After the nut fastening is completed, the bolt picking structure 8 and the nut picking structure 11 release the bolt and the nut, and the task is completed. The slider structure can be any structure that can slide up and down in the prior art. The sliding structure in this embodiment is a sliding rail and slider matching structure. The sliding rail is installed on the mounting bracket, and the slider can slide up and down along the sliding rail driven by a micro motor. The bolt picking structure is installed on the slider.

[0023] Based on the above intelligent gripper structure, this embodiment also discloses a control method for the intelligent gripper structure. The gripper structure integrates a high-resolution 3D structured light vision sensor, which is usually installed at the front end or both sides of the gripper to ensure that the three-dimensional information of the target object can be captured in real time. The installation of the sensor needs to meet the following requirements: the field of view angle needs to cover the entire operation area to ensure that the complete three-dimensional information of the target object can be captured; the sensor needs to have high resolution and high precision to generate detailed three-dimensional point cloud data; the installation position is stable to avoid inaccurate data collection caused by vibration or movement; the installation position of the sensor should avoid being blocked by the gripper or other mechanical components to ensure that the light pattern can be projected and reflected smoothly; the sensor needs to be accurately calibrated to ensure that the projected light pattern and the collected reflected information can accurately correspond. Through reasonable installation and calibration, the vision sensor can provide high-precision real-time three-dimensional data support for the intelligent gripper. To achieve target positioning, the improved iterative closest point algorithm (ICP algorithm) in this embodiment is used to register and match the point cloud data scanned in real time with the preset three-dimensional model of the wrist arm. The data frames in the optimization window are dynamically adjusted to smooth the registration result, effectively suppressing the jump problem that may occur in the single-frame registration process during the high-speed movement of the wrist arm. And innovatively, the inertial measurement unit (IMU) data is directly incorporated into the construction process of the loss function to further improve the accuracy and robustness of the registration. In addition, the size of the sliding window can be flexibly adjusted according to the specific application scenario. By optimizing the window size, the balance between real-time performance and accuracy requirements is achieved, providing strong support for the precise control of the wrist arm in the high-speed movement scenario, so as to accurately extract the spatial coordinates (X, Y, Z) and attitude angle θ of the insulator. The visual data is mapped to the manipulator control coordinate system through the coordinate transformation matrix to ensure the real-time data synchronization between the vision system and the gripper adjustment device, providing precise guidance for subsequent actions.

[0024] The control method covers dynamic calibration, closed-loop control, and fault tolerance mechanisms. During the dynamic calibration phase, the vision system periodically scans the target position at a frequency of 10 Hz, detects the small offsets of the wrist arm during the installation process, and triggers dynamic adjustment instructions for the gripper posture. The closed-loop control adopts an optimized control algorithm based on dynamic tuning of fuzzy PID parameters, combines the error signals (position and angle deviation amounts) of the vision feedback, and adjusts the cylinder output torque and the guide rail displacement in real time; uses the improved iterative closest point algorithm to register and match the point cloud data scanned in real time with the preset 3D model of the wrist arm, dynamically adjusts the data frames within the optimization window, smooths the registration results, and directly incorporates the inertial measurement unit data into the construction process of the loss function.

[0025] The real-time adjustment of the cylinder output torque and the guide rail displacement is specifically as follows: Step 1, using the friction coefficient of the clamping surface , the real-time inclination angle of the wrist arm , and the clamping force deviation as input variables; the specific quantization method for the input variables is as follows: Friction coefficient : Obtained in real time through a spectral sensor, divided into low, medium, and high grades, where the low grade , the medium grade , the high grade , and its membership function adopts a Gaussian distribution: ; Among them, is the membership function of the low grade; is the membership function of the medium grade; is the membership function of the high grade; Wrist arm inclination angle : Calculated by the vision system, and the membership function is trapezoidally distributed: ; Clamping force deviation : The difference between the target clamping force and the actual value , that is ;

[0026] Its fuzzy set is {negative large, negative small, zero, positive small, positive large}, and the membership function adopts a triangular distribution.

[0027] Step 2, dynamic adjustment of PID parameters and anti-saturation strategy: Introduce a momentum factor , weight and fuse the historical adjustment amount and the current error change, and calibrate the value through experiments to balance the real-time response and parameter stability; ; where, is the value of the scaling factor at time t+1, is the change in the scaling factor, and t is the current time step.

[0028] An anti-windup strategy is introduced. When the clamping force approaches the anti-compression threshold of the boom, the integral term is frozen to prevent overshoot caused by integral accumulation: ; where, is the control gain, is the anti-compression threshold of the boom, is the actual value of the clamping force; Step 3, construct the energy function , where , and by designing the control law to ensure that its derivative satisfies: ; where, is the derivative of the energy function, is the error change rate.

[0029] Under the anti-windup strategy, when , the integral term is frozen to avoid integral saturation, ensuring that the clamping force error asymptotically converges to zero, and finally achieving precise alignment between the gripper and the boom (positioning accuracy ≤ 0.3 mm). To cope with complex working conditions, the system designs a multi-level fault tolerance mechanism: when the 3D vision system fails due to occlusion, strong light interference or data anomalies, it immediately switches to the safe mode, uses the historical motion trajectory data interpolation to predict the target position, and starts the redundant sensor-assisted positioning to ensure the coherence and safety of the gripper movement.

[0030] The technical advantages of this method are reflected in two aspects: full-process vision guidance and high robustness. The full-process vision guidance depends on 3D vision data driving from initial positioning to dynamic adjustment, can adapt to the size differences of the booms, further improve the accuracy and robustness of registration, and solve the installation problem of non-standard booms.

[0031] To effectively suppress the possible jump problem in the single-frame registration process during the high-speed movement of the boom, this study uses dynamic adjustment to optimize the data frames within the window, smooth the registration results, and innovatively incorporates the data of the inertial measurement unit (IMU) directly into the construction process of the loss function, further improving the accuracy and robustness of registration. In addition, the size of the sliding window can be flexibly adjusted according to the specific application scenario. By optimizing the window size, the balance between real-time performance and accuracy requirements is achieved, providing strong support for the precise control of the boom in high-speed movement scenarios.

[0032] The improved Iterative Closest Point (ICP) algorithm is as follows: First, statistical outlier filtering is used to remove discrete points caused by environmental interference. Subsequently, voxel grid downsampling is employed to reduce data density, enhancing computational efficiency while preserving key geometric features. Regarding the surface characteristics of the wrist arm insulator, the normal vector distribution or curvature features can be further extracted to construct local feature descriptors, providing robustness guarantees for subsequent matching.

[0033] The core objective of the ICP algorithm is to find the rotation matrix R and translation vector s. Let the initial transformation matrix be , define the sliding window size as N, indicating the joint optimization of the pose parameters of the nearest N frames, and the initial spatio-temporal loss weight is , which is dynamically adjusted based on the motion speed subsequently.

[0034] During iterative calculation (the k-th iteration), for each point , search for the bidirectional nearest neighbor point in the reference point cloud Q, and introduce the dynamic distance threshold : ; Among them, , is the initial threshold, is the attenuation coefficient, and k is the iteration number.

[0035] To suppress the influence of outlier points, introduce the weighted centroid based on the matching confidence and , where ; Among them, i is the three-dimensional coordinate of the i-th point, from the source point cloud p, N is the total number of coordinates, is the weight coefficient of the i-th point in the k-th iteration, and j is the three-dimensional coordinate of the j-th point, from the target point cloud q.

[0036] Secondly, jointly construct the covariance matrix through multiple frames of data within the sliding window to reduce computational redundancy: .

[0037] Perform weighted SVD decomposition on the covariance matrix , where is the diagonal weight matrix, and the elements are , and update the rotation matrix and translation vector: ; .

[0038] Define the spatio-temporal loss function , which combines geometric error and motion consistency to achieve spatio-temporal joint optimization of the sliding window.

[0039] ; Among them, is deduced from IMU data, is the number of points participating in the matching at time t.

[0040] Weight , Dynamic adjustment: ; Among them, is the current motion speed, is the adjustment coefficient.

[0041] Solve the minimization problem of the spatio-temporal loss function and update the transformation matrix of all frames within the window , .

[0042] Comprehensively judge convergence based on the following conditions: Mean square error change: ; Maximum number of iterations: ; Stability of the transformation matrix: ; Among them is the convergence threshold, is the maximum number of iterations.

[0043] If convergence is judged, output the optimized transformation matrix of the current frame , if not converged, predict the initial pose of the next frame using IMU data: .

[0044] The improved ICP algorithm, through spatio-temporal joint optimization of multiple frames in the sliding window and IMU data-driven non-linear motion prediction, suppresses the jump of single-frame registration, is more suitable for the high-speed motion scenario of the wrist arm, the size N of the sliding window is adjustable, further improving the accuracy and robustness of registration, and providing strong support for the precise control of the wrist arm in the high-speed motion scenario.

[0045] Set a fault tolerance mechanism during the control process. When the vision system fails due to occlusion, strong light interference or data anomaly, switch to the safe mode, interpolate and predict the target position using historical motion trajectory data, and start redundant sensor-assisted positioning.

[0046] The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present invention, and such modifications or equivalent substitutions should also be regarded as falling within the protection scope of the present invention.

Claims

1. A method for controlling an intelligent gripper structure for installing a contact network arm, characterized in that: It includes a dynamic calibration stage and a closed-loop control stage. In the dynamic calibration stage, the visual system periodically scans the target position, detects the offset of the wrist arm during the installation process, and triggers a dynamic adjustment instruction of the gripper posture; The closed-loop control stage adopts an optimization control algorithm based on dynamic adjustment of fuzzy PID parameters, combined with the position and angle deviation feedback from the visual system, to adjust the cylinder output torque and guide rail offset in real time; an improved iterative closest point algorithm is used to align and match the real-time scanned point cloud data with the preset arm 3D model, dynamically adjust the data frame within the optimization window, smooth the alignment results, and directly incorporate the inertial measurement unit data into the loss function construction process; A fault-tolerant mechanism is set up during the control process. When the visual system fails due to occlusion, strong light interference or data anomaly, it switches to safe mode, uses historical motion trajectory data to interpolate and predict the target position, and starts redundant sensors to assist in positioning.

2. The intelligent gripper structure control method for contact network arm installation according to claim 1 is characterized in that: The real-time adjustment of the cylinder output torque and the guide rail displacement is specifically as follows: Step 1: Use the friction coefficient of the clamping surface 、Real-time inclination of wrist and arm and the clamping force deviation e are input variables; Step 2: Dynamic adjustment of PID parameters and anti-saturation strategy: Introducing momentum factor , weighted fusion of historical adjustment and current error change, and calibrated by experiments value, balancing real-time response and parameter stability; ； in, is the value of the proportional factor at time t+1, is the change of the proportional factor, and t is the current time step; The anti-saturation strategy is introduced. When the clamping force approaches the arm compression threshold, the integral term is frozen: ； in, To control the gain, is the wrist-arm pressure resistance threshold, is the actual value of the clamping force; Step 3: Construct energy function ,in , by designing the control law to ensure that its derivative satisfies: ; in, is the derivative of the energy function, is the error change rate.

3. The intelligent gripper structure control method for contact network arm installation according to claim 2 is characterized in that: The specific quantification method of the input variables in step 1 is: Friction coefficient : It is acquired in real time through the spectral sensor and divided into three levels: low, medium and high. , mid-range ,upscale , and its membership function adopts Gaussian distribution: ； in, is the low-grade membership function; is a mid-range membership function; is a high-end membership function; Arm inclination : Calculated by the visual system, the membership function is a trapezoidal distribution: ； Clamping force deviation : Target holding force With actual value The difference, that is ; Its fuzzy set is {negative large, negative small, zero, positive small, positive large}, and the membership function adopts triangular distribution.

4. The intelligent gripper structure control method for contact network arm installation according to claim 1 is characterized in that: The improved iterative closest point algorithm is specifically as follows: by rotating the matrix R and translating the vector s, the initial spatiotemporal loss weight is set and , initial transformation matrix , define the sliding window size as N, which means jointly optimizing the pose parameters of the latest N frames; During iterative calculation, for each point , search for the bidirectional nearest neighbor points in the reference point cloud Q , and introduce a dynamic distance threshold : ; in, , is the initial threshold, is the attenuation coefficient, k is the number of iterations; In order to suppress the influence of outliers, a matching confidence The weighted centroid of and , is the noise standard deviation; Construct the covariance matrix by jointly constructing multiple frames of data in the sliding window , for the covariance matrix Perform weighted SVD decomposition and update the rotation matrix and translation vector; Define the spatiotemporal loss function , integrating geometric error and motion consistency to achieve spatiotemporal joint optimization of sliding windows; Weighting of spatiotemporal loss and Make dynamic adjustments; Solve the minimization problem of the spatiotemporal loss function and update the transformation matrix of all frames in the window ; Based on the change of mean square error, maximum number of iterations, and stability of transformation matrix, whether it converges is determined. If converged, the optimized transformation matrix of the current frame is output. , If it does not converge, use IMU data to predict the initial pose of the next frame.

5. The intelligent gripper structure control method for contact network arm installation according to claim 4, characterized in that: The weighted centroid and The calculation formula is: ； Where i is the 3D coordinate of the i-th point from the source point cloud p, N is the total number of coordinates, is the weight coefficient of the i-th point in the k-th iteration, and j is the three-dimensional coordinate of the j-th point, which comes from the target point cloud q.

6. The intelligent gripper structure control method for contact network arm installation according to claim 4, characterized in that: The covariance matrix The calculation formula is: ； Spatiotemporal loss function The calculation formula is: ; ; in, Calculated from IMU data, is the number of points involved in matching at time t, is the current movement speed, is the adjustment coefficient.

7. An intelligent clamping claw structure for installing a contact network arm, characterized in that: A method for controlling an intelligent clamping structure for installation of a contact network arm as described in any one of claims 1 to 6, comprising a visual sensor, a back plate, a clamping structure, and a bolt and nut picking structure, wherein the visual sensor is mounted on the back plate, and at least three groups of clamping structures are mounted on the back plate, wherein the clamping structures are arranged in a triangle, wherein one group of the clamping structures is mounted on the back plate via an adjustable slide rail, and the bolt and nut picking structure is mounted on the side of the clamping structure corresponding to one end of the contact network arm where the bolt and nut need to be installed.

8. The intelligent clamping claw structure for installing a contact network arm according to claim 7, characterized in that: The bolt and nut picking structure includes a mounting frame, a bolt picking structure, and a nut picking structure. The bolt picking structure is installed on the mounting frame through a sliding structure. The nut picking structure corresponds to the bolt picking structure to ensure the matching of the bolt and nut. Both the bolt picking structure and the nut picking structure are picked up through electromagnetic force.

9. The intelligent clamping claw structure for installing a contact network arm according to claim 7, characterized in that: The adjustable slide rail includes a linear slide rail and an arcuate slide rail. The linear slide rail is installed on the back plate, and the arcuate slide rail is installed on the linear slide rail. The arcuate slide rail can slide up and down along the linear slide rail. The clamping structure is installed on the arcuate slide rail and can slide along the arcuate slide rail.

Images