End position dynamic compensation method based on fusion of visual servoing and force feedback

Abstract

The application relates to the technical field of robot control, and discloses a terminal position dynamic compensation method based on fusion of visual servoing and force feedback, which comprises the following steps: collecting a top view image, segmenting an irregular residual hole contour, and calculating a dynamic alignment center coordinate; then, a reverse conical spiral descending trajectory is generated in combination with a lateral force vector to guide a fiber tail fiber connector into a guide port; in the insertion stage, the curvature radius characteristic of a force-displacement curve resistance peak point is used to determine the authenticity of bottom touching; for the case that the bottom touching is determined to be false, the side projection chord length is used to deduce an under-insertion depth, and a high-frequency micro-oscillation compensation insertion is driven until a real buckle locking feature is captured; the application realizes accurate alignment and reliable positioning of fiber plug-in and plug-out actions under complex working conditions such as shielding and mechanical wear.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, and more specifically, to a method for dynamic end-effector pose compensation based on the fusion of visual servoing and force feedback. Background Technology

[0002] With the continuous expansion of data centers and communication equipment rooms, the need for automated maintenance of fiber optic distribution frames (ODFs) is becoming increasingly urgent. In automated insertion and removal operations, robotic arms typically need to integrate machine vision and multi-dimensional force sensing to achieve high-precision positioning and smooth insertion of fiber optic connectors in complex environments.

[0003] Patent CN121374654B discloses a method and system for assembling cables for a robotic arm. This technology fits the cable centerline by acquiring point cloud data of the working environment and plans the insertion path according to the relative pose matrix. It uses the rate of change of insertion force and the end-effector pose error as the criteria for determining whether the assembly is complete. Patent CN117245651B discloses a method, device, equipment, and storage medium for controlling the insertion and removal of a robotic arm. It uses a vision sensor to acquire image information of the target object for preliminary positioning, searches for the target hole through feedback information from a force sensor during movement, and finally performs insertion and removal based on a compliant force control algorithm.

[0004] However, in actual maintenance scenarios involving dense fiber optic patch panels, the tightly packed cables can easily cause lateral compression and obstruction of the target slot, leading to geometric topological distortion of the originally standard rectangular slot entrance. This causes the center of the actually usable insertion aperture to deviate from the physical center of the slot. In this case, if alignment is based solely on a fixed visual template or physical coordinates, the connector edge is prone to hard friction with surrounding cables, and may even induce uncontrollable lateral slippage under random lateral forces, causing the connector to jam in the gap between adjacent cables. Furthermore, in older cabinets, the dry friction resistance generated by aging, deformed, or dust-accumulated springs in the slots often creates a false force peak in the middle of the insertion section that closely matches the actual locking characteristics. Because the end grippers physically obstruct the vertical insertion direction, the vision system cannot directly observe the positioning status in the blind zone. If the control system relies solely on the slope of the peak descent of the force curve for judgment, it is very easy to misjudge the insertion as complete due to this "false bottoming out" signal, resulting in the cable being in an unlocked, loose connection state, which seriously affects the stability of the communication link and the safety of equipment operation. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, this invention provides a dynamic end-effector pose compensation method based on the fusion of visual servoing and force feedback. By identifying the irregular hole contours after pressure distortion, the dynamic alignment center is determined. Combined with the inverted conical spiral trajectory generated by the lateral force vector, the risk of hole center offset and lateral slippage caused by dense cable compression is eliminated. The authenticity is determined by using the neighborhood curvature radius of the resistance peak point of the force-displacement curve, which can accurately remove dry friction interference signals from the geometric feature level. With the depth inversion mechanism of the side projection chord length and high-frequency micro-oscillation compensation, the problem of positioning quantization and friction bottleneck breakthrough in the vertical visual blind zone is solved, ensuring the reliability of physical locking in the complex force environment under dry friction interference.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A dynamic end-effector pose compensation method based on the fusion of visual servoing and force feedback includes:

[0008] A top-view image of the target fiber optic slot and its surrounding area is acquired. The polygonal outline of the irregular remaining hole after being blocked by the surrounding cables is segmented from the top-view image. The dynamic alignment center coordinates are obtained based on the polygonal outline of the irregular remaining hole.

[0009] The lateral force vector is obtained based on the dynamic alignment center coordinates. Based on the lateral force vector, an inverted conical spiral descent trajectory is generated. The gripper of the robotic arm is controlled to send the fiber optic pigtail connector into the guide port of the target fiber optic slot along the inverted conical spiral descent trajectory.

[0010] After the fiber optic pigtail connector enters the guide port of the target fiber slot, a force-displacement curve is plotted. When the force-displacement curve shows a resistance peak point, the radius of curvature of the curve in the neighborhood of the resistance peak point is calculated. The radius of curvature of the curve is compared with a preset radius of curvature threshold to determine the authenticity of bottoming out.

[0011] When a false bottoming out is determined, the side projection chord length is obtained, the under-insertion depth is inverted based on the side projection chord length, and high-frequency micro-oscillation compensation insertion is performed according to the under-insertion depth until the force-displacement curve shows the true locking characteristics.

[0012] The method for obtaining the polygonal outline of the irregular remaining holes includes:

[0013] Semantic segmentation is performed on the top view image to distinguish the passable area of ​​the target fiber optic slot opening from the obstruction area of ​​the surrounding cables. The outer boundary of the passable area is extracted and fitted into an irregular residual hole polygonal contour. The coordinates of each vertex of the irregular residual hole polygonal contour are stored.

[0014] The method for obtaining the dynamic alignment center coordinates includes:

[0015] Read the nominal length and nominal width values ​​of the cross-section of the fiber optic pigtail connector, calculate the aspect ratio of the cross-section of the fiber optic pigtail connector, solve for the maximum inscribed rectangle within the irregular remaining hole polygonal outline with the aspect ratio of the cross-section of the fiber optic pigtail connector as the constraint condition, and extract the center coordinates of the maximum inscribed rectangle as the dynamic alignment center coordinates.

[0016] The method for obtaining the lateral force vector includes:

[0017] When the length of the short side of the largest inscribed rectangle meets the safety margin requirement, the robotic arm is controlled to translate to directly above the dynamic alignment center coordinates; from directly above the dynamic alignment center coordinates, the Z-axis descent is executed. During the Z-axis descent, the lateral force components in the X-axis direction and the Y-axis direction are detected in real time by a six-dimensional force sensor, and the lateral force components in the X-axis direction and the Y-axis direction are combined into a lateral force vector.

[0018] The condition for meeting the safety margin requirement is that the length of the short side of the largest inscribed rectangle is greater than or equal to the sum of the nominal width of the cross-section of the fiber optic pigtail connector and the preset safety margin.

[0019] The method for generating the inverted conical spiral descent trajectory includes:

[0020] The resultant force amplitude of the lateral force vector is compared with a preset compliance threshold. When the resultant force amplitude of the lateral force vector exceeds the preset compliance threshold, the gripper of the control robot arm generates a horizontal translation component along the normal direction of the lateral force vector. The displacement generated by the Z-axis actuator continuously descending in the vertical direction is recorded as the Z-axis descent component. The horizontal translation component and the Z-axis descent component are superimposed to form an inverted conical spiral descent trajectory.

[0021] The method for determining whether to insert the fiber optic pigtail connector into the guide port of the target fiber optic slot includes:

[0022] The fiber optic pigtail descends along an inverted conical spiral trajectory until it enters the guide port of the target fiber optic slot at the bottom of the fiber optic pigtail connector. When a rigid wall constraint resistance is detected in the Z-axis direction and the resultant force amplitude of the lateral force vector drops to near zero, it is determined that the fiber optic pigtail connector has entered the guide port of the target fiber optic slot. Near zero means that the fluctuation value drops to less than the preset lower limit threshold of the lateral force.

[0023] The method for plotting force-displacement curves includes:

[0024] After the fiber optic pigtail connector enters the guide port of the target fiber slot, the Z-axis actuator is controlled to continue to press down and advance at a constant speed in the vertical direction, and the cumulative downward displacement of the Z-axis and the insertion resistance in the Z-axis direction are recorded. The force-displacement curve is plotted point by point with the cumulative downward displacement of the Z-axis as the abscissa and the insertion resistance in the Z-axis direction as the ordinate.

[0025] The method for calculating the radius of curvature of the curve in the neighborhood of the resistance peak point includes: extracting curve data segments within a preset window width range before and after the resistance peak point, performing second derivative calculation on the curve data segments, and obtaining the radius of curvature of the curve in the neighborhood of the resistance peak point.

[0026] The methods for determining whether a bottom has been reached include:

[0027] When the radius of curvature of the curve is less than the preset radius of curvature threshold, the resistance peak point is determined to correspond to a real latching event; when the radius of curvature of the curve is greater than or equal to the preset radius of curvature threshold, the resistance peak point is determined to correspond to a false bottoming event, the downward action of the Z-axis actuator is frozen, and the current cumulative downward displacement of the Z-axis is marked as the false bottoming frozen displacement.

[0028] The method for obtaining the side projection chord length includes:

[0029] In the pseudo-bottoming and freezing state, the side image of the fiber optic pigtail connector is acquired from a preset oblique viewing angle. The trapezoidal projection contour of the fiber optic pigtail connector exposed outside the target fiber slot is identified from the side image. The side pixel length extending along the height direction of the fiber optic pigtail connector in the trapezoidal projection contour is extracted and recorded as the side projection chord length.

[0030] The method for inverting under-interpolation depth based on side-projected chord length includes:

[0031] An insertion depth mapping model is constructed, and the side projection chord length is input into the insertion depth mapping model to obtain the under-insertion depth of the fiber optic pigtail connector from the actual latching position.

[0032] The method for performing high-frequency micro-oscillation compensation insertion includes:

[0033] Based on the under-insertion depth control, the Z-axis actuator performs high-frequency micro-oscillation with a preset micro-oscillation amplitude and preset micro-oscillation frequency on the basis of the false bottoming-freezing displacement. When the under-insertion depth approaches zero and the radius of curvature of the curve in the neighborhood of the newly appeared resistance peak point is less than the preset radius of curvature threshold, it is determined that the fiber optic pigtail connector has reached the true locking position, the high-frequency micro-oscillation stops, the force is recovered and the clamp is released to complete the insertion operation. Approaching zero means that the monitored current under-insertion depth is less than the preset under-insertion depth convergence threshold.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] This invention effectively eliminates the impact of hole topological distortion caused by dense cable packing on positioning accuracy by extracting the dynamic alignment center from the polygonal contour of irregular holes, making the alignment process no longer dependent on idealized physical center coordinates. The combination of the inverted conical spiral descent trajectory and the lateral force vector endows the end effector with active compliance and deviation correction capabilities during the guide port entry stage, avoiding component damage caused by hard friction. Utilizing the radius of curvature of the resistance peak neighborhood of the force-displacement curve for authenticity determination, it can distinguish between instantaneous dry friction interference and the actual elastic locking physical process from a geometric feature level, solving the force feedback illusion caused by slot aging. The side projection chord length inversion mechanism bypasses the physical obstruction constraint of the gripper on the vertical viewing angle, achieving indirect quantitative perception of blind zone depth; combined with high-frequency micro-oscillation compensation, it dynamically reduces the equivalent friction coefficient of the contact interface, enabling the connector to overcome friction resistance bottlenecks and reach the predetermined insertion depth. The overall solution, through deep coupling of visual contour quantification and force curve geometric features, ensures the reliability of the closed-loop operation in complex force perception environments with visual obstruction and dry friction interference. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart of the end-effector pose dynamic compensation method based on the fusion of visual servoing and force feedback provided in an embodiment of the present invention.

[0038] Figure 2 This is a top-down view of an obstructed scene of a fiber optic slot array provided in an embodiment of the present invention;

[0039] Figure 3 A top-view image acquisition diagram of the AI ​​vision-assisted system provided in an embodiment of the present invention;

[0040] Figure 4 A schematic diagram of a region polygon contour fitting provided for an embodiment of the present invention;

[0041] Figure 5 This is a flowchart for determining the authenticity of bottoming out, provided in an embodiment of the present invention.

[0042] Figure 6 A schematic diagram comparing the force-displacement curves of actual latch engagement and false bottoming out, provided in an embodiment of the present invention;

[0043] Figure 7This invention provides a process for determining the authenticity of a bottoming-out point in an embodiment of the invention.

[0044] Figure 8 This is a functional block diagram of an end-effector pose dynamic compensation system based on the fusion of visual servoing and force feedback, provided in an embodiment of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1:

[0047] Please see Figure 1 As shown, this embodiment provides a method for dynamic end-effector pose compensation based on the fusion of visual servoing and force feedback, including:

[0048] Step S10: Acquire a top-view image of the target fiber optic slot and its surrounding area; segment the polygonal outline of the irregular remaining holes after being blocked by surrounding cables from the top-view image; obtain the dynamic alignment center coordinates based on the polygonal outline of the irregular remaining holes; obtain the lateral force vector based on the dynamic alignment center coordinates; generate an inverted conical spiral descent trajectory based on the lateral force vector; and control the gripper of the robotic arm to send the fiber optic pigtail connector into the guide port of the target fiber optic slot along the inverted conical spiral descent trajectory.

[0049] Specifically, step S10 involves a matrix-arranged array of fiber optic slots on the chassis of a high-density fiber optic distribution frame cabinet. (See [link to relevant documentation]). Figure 2 This is a top-view diagram of an obstructed scenario for the fiber optic slot array provided in an embodiment of this application. Figure 2 The diagram illustrates adjacent slots arranged in a matrix, the target fiber optic slot to be inserted, and obstructing cables extending from the adjacent slots. When the robotic arm needs to insert a fiber optic pigtail connector into a target fiber optic slot, most of the adjacent slots are usually already full of pigtails. The numerous inserted pigtail cables exhibit irregular bending and stacking above their respective slot openings due to their own flexible rebound, gravity, and mutual compression between adjacent cables. Figure 2The presented operational scenario involves cable segments intruding from the side into the vertical channel space directly above the target fiber optic slot. This causes the originally standard rectangular slot entrance to appear as an irregularly shaped channel, cut by the cables, when viewed from above. In traditional robotic arm control schemes, the system directly reads the factory-specified physical center coordinates of the target fiber optic slot based on pre-stored cabinet slot layout drawings and controls the gripper to align with these coordinates and press down vertically. However, due to the cable intrusion, the geometric center of the actual passable area deviates from the nominal physical center coordinates. Forcibly pressing down according to the nominal coordinates will cause the edge of the fiber optic pigtail connector to make hard contact with the surrounding intruding cables, generating lateral forces of unpredictable direction and magnitude. This can range from causing the fiber optic pigtail connector to slide off the axis of the target fiber optic slot to jamming the connector into the gap between adjacent cables, preventing further advancement, or even causing irreversible physical damage to the already inserted adjacent pigtails. Step S10 uses a combination of three techniques—visual segmentation and reconstruction of the real topology of the area, computational geometry to solve for the optimal alignment position, and force-guided flexible spiral descent through the cable compression area—to safely deliver the fiber optic pigtail connector from the messy cable area above the target fiber optic slot to the rigid guide port of the target fiber optic slot. This lays the foundation for entry positioning in the subsequent step S20, which involves snap-locking and bottom-touching verification.

[0050] Further, step S10 includes:

[0051] Step S11: Determine the nominal physical center coordinates of the target fiber optic slot, control the robotic arm to move directly above the nominal physical center coordinates, and call the AI ​​vision-assisted system to collect a top-down view image of the target fiber optic slot and its surrounding area.

[0052] In step S11, see Figure 3 This is a top-view image acquisition schematic diagram of the AI ​​vision assistance system provided in this application embodiment. Figure 3The diagram illustrates the AI ​​vision assistance system's camera, the camera's optical axis, the gripper mounted at the bottom of the Z-axis actuator, the preset standby height between the gripper and the cabinet chassis, the nominal physical center coordinates of the cabinet chassis and the target fiber optic slot, and the nominal physical center coordinates of the target fiber optic slot derived from a pre-stored cabinet slot layout parameter table. This cabinet slot layout parameter table is provided by the manufacturer and stored in the system memory at the time of cabinet shipment, based on the slot array's row and column numbers, row spacing, and column spacing. Each slot corresponds to a fixed two-dimensional coordinate value based on the robotic arm's working coordinate system in the parameter table. The robotic arm's working coordinate system is defined using a right-handed Cartesian coordinate system: the X-axis is the first horizontal direction along the surface of the cabinet chassis, parallel to the row arrangement direction of the fiber optic slot array; the Y-axis is the second horizontal direction along the surface of the cabinet chassis, parallel to the column arrangement direction of the fiber optic slot array. The X-axis and Y-axis are perpendicular to each other in the horizontal plane and together form a horizontal reference plane parallel to the surface of the cabinet chassis; the Z-axis is the vertical direction perpendicular to the surface of the cabinet chassis, with the positive direction being upwards away from the cabinet chassis. After receiving the insertion operation command, the robotic arm retrieves the nominal physical center coordinates corresponding to the row and column numbers of the target fiber optic slot from the cabinet slot layout parameter table. Then, it controls the X-axis guide rail drive module and the Y-axis guide rail drive module to coordinate their actions, translating the gripper mounted at the bottom of the Z-axis actuator to directly above the nominal physical center coordinates. Figure 3 The presented structural layout places the grippers directly above the nominal physical center coordinates, precisely corresponding to the target fiber optic slot position on the chassis below. At this point, the Z-axis actuator has not yet initiated its descent, and the grippers remain at a distance from the chassis. Figure 3 The marked position of the preset standby height. The value of the preset standby height must ensure that the gripper body does not enter the top-down observation field of view of the AI ​​vision-assisted system, so as not to obstruct the target fiber optic slot opening area, such as... Figure 3 As shown, the camera's optical axis points downwards vertically. The preset standby height ensures the gripper won't enter the camera's field of view and obstruct it, while avoiding excessive height to prevent an excessively long Z-axis descent, which would reduce work efficiency. For example, the preset standby height is approximately 30mm to 50mm. While the gripper is still at the preset standby height, the system invokes the AI ​​vision assistance system mounted above the robotic arm, such as... Figure 3As shown, the camera optical axis of the AI ​​vision assistance system points vertically downwards, capturing a full-resolution image of the target fiber optic slot and its surrounding area from directly above, generating a top-down view. The image acquisition is scheduled before the Z-axis actuator begins its descent, rather than during the descent itself, because the gripper itself is a rectangular block structure with a certain outer dimension. Once the Z-axis actuator begins its descent, the gripper will gradually obscure the visual observation area directly below. The effective observation window of the AI ​​vision assistance system continuously shrinks as the gripper descends deeper, until it is completely obscured. Images acquired while the gripper is obscured will lack crucial information about the target fiber optic slot opening area. Therefore, completing image acquisition when the gripper is at a preset standby height and does not obscure the target area allows for the acquisition of complete top-down field-of-view information about the target fiber optic slot opening area, including the outline of the passable area at the slot opening and the spatial distribution of the inserted pigtail cables extending from adjacent slots towards the target fiber optic slot. The top-view image acquired in step S11 serves as the raw input data for semantic segmentation in step S12. Without the image acquisition step in S11, subsequent step S12 would be unable to obtain the occlusion status information of the target fiber optic slot opening area, the irregular remaining hole polygonal outline could not be generated, and the maximum inscribed rectangle solution in step S13 would lack a constraint domain input. The entire dynamic alignment process would fail to start due to the broken information source. Step S11 captures and transmits the complex cable occlusion distribution information in the actual cabinet environment, which changes with each operation, as a complete top-view image to the subsequent digitization process, ensuring that the analysis and control in step S10 do not rely on any prior assumptions about cable distribution.

[0053] Step S12: Perform semantic segmentation on the top view image, distinguish the passable area of ​​the target fiber optic slot opening from the obstruction and coverage area of ​​the surrounding cables, extract the outer boundary of the passable area and fit it into the polygonal contour of the irregular remaining hole, and store the coordinates of each vertex of the polygonal contour of the irregular remaining hole.

[0054] See Figure 4 This is a schematic diagram of a region polygon contour fitting provided in an embodiment of this application. Figure 4The image illustrates the original rectangular boundary of the target fiber optic slot, the area obstructed by the cable, the passable area not obstructed by the cable, the polygonal outline of the irregular remaining hole generated by fitting, and the coordinates of the vertices of the polygon constituting the outline. Step S12 performs semantic segmentation processing on the top-view image output in step S11, classifying each pixel in the image into two semantic categories. The AI ​​vision-assisted system pre-deploys a semantic segmentation network model trained under supervised learning. The training dataset comes from a large number of top-view images of the slot opening area collected in real cabinet environments with different service lives and cable insertion / removal densities. All pixels in each training image are manually labeled into two categories: such as... Figure 4 The presented region segmentation results classify pixels belonging to the passable area of ​​the target fiber optic slot opening that is not yet covered by cables as foreground pixels, and pixels belonging to the surrounding cable-covered areas and non-passage structures such as the slot frame as background pixels. The trained and converged semantic segmentation network model receives the top-view image input and outputs a confidence score for each pixel in the image, classifying it into either the foreground or background category. Based on a preset confidence threshold, the system classifies all pixels into a passable region set and a non-passable region set. Subsequently, connected component extraction is performed on the passable region set to identify the largest connected pixel region within the target fiber optic slot opening area. A boundary tracking algorithm is then performed on this connected region to obtain the closed outer boundary pixel sequence of the passable region. Directly using pixel-level jagged boundaries for subsequent geometric calculations would lead to decreased solution accuracy and increased computation time. Therefore, the system performs polygon fitting on the boundary pixel sequence, employing the Douglas-Peucker polyline simplification algorithm to approximate the curved outer boundary as a closed polygon formed by several line segments connected end-to-end. Figure 4 The irregular remaining hole polygon outline is marked in the image. The Douglas-Peucker algorithm balances the relationship between the number of polygon vertices and the contour reconstruction accuracy by controlling the simplification tolerance parameter: a smaller simplification tolerance retains more vertices and the polygon is closer to the original boundary, but the computational complexity is higher; a larger simplification tolerance results in fewer vertices but increases contour deviation. The value of the simplification tolerance is determined by comparing the deviation between the maximum inscribed rectangle area output in subsequent step S13 at different tolerance levels during the calibration stage and the reference area solved using the original pixel-level boundary. The maximum tolerance value corresponding to a deviation less than 5% of the cross-sectional area of ​​the fiber optic pigtail connector is selected to minimize the number of polygon vertices while meeting accuracy constraints. The system will fit the obtained... Figure 4The coordinates of each vertex of the irregular remaining hole polygon outline marked are transformed from the image pixel coordinate system to the physical coordinates in the working coordinate system of the robotic arm. The mapping matrix on which the transformation is based is obtained and stored by the calibration board calibration program during the initial system installation. The transformed vertex coordinates are written into the system memory for step S13 to read. Step S12 compresses the complex, pixel-level distributed occlusion information in the top view image into a polygonal mathematical object described by a finite number of physical coordinate vertices. This allows the subsequent step S13 to run deterministic computational geometry algorithms for accurate spatial reasoning, rather than performing a costly and unstable pixel-by-pixel search in the original image pixel space. Without the semantic segmentation and polygon fitting steps of step S12, step S13 will not be able to obtain deterministic constraint input describing the boundary of the passable region. The solution of the maximum inscribed rectangle will lack a defined constraint domain. The acquisition of the dynamic alignment center coordinates will degenerate into the direct adoption of the nominal physical center coordinates. The adaptive capability of the entire step S10 for cable occlusion distortion will be completely lost.

[0055] Step S13: Read the nominal length and nominal width values ​​of the cross-section of the fiber optic pigtail connector, calculate the aspect ratio of the cross-section of the fiber optic pigtail connector, solve for the maximum inscribed rectangle within the irregular remaining hole polygonal outline with the aspect ratio of the cross-section of the fiber optic pigtail connector as the constraint condition, and extract the center coordinates of the maximum inscribed rectangle as the dynamic alignment center coordinates.

[0056] Step S13 reads the nominal length and nominal width values ​​of the fiber optic pigtail connector cross-section from the pre-stored fiber optic pigtail connector specification parameter table in the system, and divides the nominal length value by the nominal width value to obtain the aspect ratio of the fiber optic pigtail connector cross-section. For example, for an SC type fiber optic pigtail connector, the nominal length value of the cross-section is approximately 9.0 mm, the nominal width value is approximately 4.5 mm, and the aspect ratio of the fiber optic pigtail connector cross-section is approximately 2.0; for an LC type fiber optic pigtail connector, the nominal length value of the cross-section is approximately 6.3 mm, the nominal width value is approximately 3.0 mm, and the aspect ratio of the fiber optic pigtail connector cross-section is approximately 2.1. Using the vertex coordinates of the irregular remaining hole polygon contour output and stored in memory in step S12 as the boundary of the constraint domain, and the aspect ratio of the fiber optic pigtail connector cross-section as the equivalent constraint condition for the aspect ratio of the rectangle, a computational geometry search algorithm is run to search for a rectangle within the region enclosed by the irregular remaining hole polygon contour that simultaneously satisfies the following two conditions: all four sides and four vertices of the rectangle are located inside the irregular remaining hole polygon contour or exactly on the boundary, and the ratio of the longer side to the shorter side of the rectangle is equal to the aspect ratio of the fiber optic pigtail connector cross-section. Among all candidate rectangles that satisfy the above two conditions, the one with the largest area is selected as the maximum inscribed rectangle. The search algorithm is implemented using a combination of rotational scanning and binary width search. The azimuth of the candidate rectangle is rotated incrementally within a range of 0° to 180° (e.g., the preset angle increment is approximately 1° to 5°). At each azimuth, the maximum rectangle width that can be accommodated under a given aspect ratio constraint is checked along each edge of the irregular remaining hole polygon outline. A binary search is used to narrow the width range until convergence to the required accuracy. The width, length, and center coordinates of the candidate rectangle with the largest area at that azimuth are recorded. After traversing all azimuths, the largest area is selected from all records as the final result. For example, when the irregular remaining hole polygon outline is approximately a parallelogram and its major axis is at an angle of approximately 30° to the long side of the fiber optic pigtail connector cross-section, the search algorithm finds the largest inscribed rectangle with the largest area near the 30° azimuth angle. Its long side is along the major axis of the parallelogram, and its short side is along the minor axis of the parallelogram.

[0057] Extract the center coordinates of the largest inscribed rectangle and set them as the dynamic alignment center coordinates, replacing the nominal physical center coordinates of the target fiber optic slot as the alignment target for the grippers in the horizontal plane. The choice of the center coordinates of the largest inscribed rectangle, rather than the geometric centroid of the irregular remaining hole polygon outline, as the alignment target is based on the following physical reasoning: the geometric centroid of an irregular polygon is the first-order centroid of its area distribution. When the polygon is elongated, asymmetrical, or contains deep recesses, the geometric centroid may fall within a narrow section near the edge of the recess inside the polygon. The available channel width at this location along a certain direction may be smaller than the nominal width of the fiber optic pigtail connector cross-section. Using this location as the alignment target and applying downward pressure will directly cause the edge of the fiber optic pigtail connector to strike the cable at the polygon boundary; while the largest inscribed rectangle... The center coordinates of the shape are located in the center of the largest inscribed area of ​​the rectangle. The length of the short side of the largest inscribed rectangle represents the minimum unobstructed spacing available in the width direction around the center point, and the length of the long side of the rectangle represents the minimum unobstructed spacing in the length direction. The spacing in both directions has reached the maximum accommodative level within the constraint domain. When the fiber optic pigtail connector is lowered vertically with this center coordinate as the axis, its cross-sectional outer contour maintains a gap between the boundary of the irregular remaining hole polygon contour in both orthogonal directions, which geometrically ensures the collision-free safety of the starting section of the lowering path.

[0058] Step S13 also includes a safety margin verification logic: comparing the sum of the shorter side length of the largest inscribed rectangle and the nominal width of the fiber optic pigtail connector cross-section plus a preset safety margin. The preset safety margin is set based on the maximum lateral offset that the fiber optic pigtail connector may experience due to manufacturing tolerances of the grippers and eccentricity of the gripping force point while held in the gripper state. This value is obtained by repeatedly performing clamping, releasing, and measuring deviation calibration tests on a standard test bench, and the upper limit of the statistical deviation distribution is used as the preset safety margin. For example, the preset safety margin is approximately 0.3 mm. When the shorter side length of the largest inscribed rectangle is greater than or equal to the sum of the nominal width and the preset safety margin, it is determined that the current gap can accommodate the cross-section of the fiber optic pigtail connector in the width direction and leave room to cope with clamping deviations. The safety margin is deemed to meet the requirements, and then the process proceeds to step S14. When the shorter side length of the largest inscribed rectangle is less than the sum of the above, it is determined that the current cable obstruction level is insufficient to safely accommodate the fiber optic pigtail connector. The insertion operation is then stopped, and a cable clearing request is generated. After the obstructing cables around the target fiber optic slot are cleared and dispersed, the process returns to step S11. The safety margin check logic enables the system to proactively stop when the obstruction is too severe, rather than blindly attempting to force it down, thus avoiding irreversible physical damage caused by rigid collisions between the fiber optic pigtail connector and surrounding inserted cables. The irregular remaining hole polygon outline output in step S12 provides a precise constraint domain boundary for the computational geometry search in step S13. The search algorithm running within this constraint domain in step S13 transforms the irregular boundary information of the polygon into a rectangular geometric object with clear center coordinates and size parameters. Together, these two elements condense the complex and random cable obstruction distribution information in the top view image into a two-dimensional coordinate quantity that can directly drive the robotic arm to perform precise translation.

[0059] Step S14: When the length of the short side of the largest inscribed rectangle meets the safety margin requirement, control the robotic arm to translate to directly above the dynamic alignment center coordinates; start the Z-axis descent from directly above the dynamic alignment center coordinates. During the Z-axis descent, the lateral force components in the X-axis direction and the Y-axis direction are detected in real time by a six-dimensional force sensor. The lateral force components in the X-axis direction and the Y-axis direction are combined into a lateral force vector, and the resultant force amplitude of the lateral force vector is calculated.

[0060] The condition for meeting the safety margin requirement is that the length of the short side of the largest inscribed rectangle is greater than or equal to the sum of the nominal width of the cross-section of the fiber optic pigtail connector and the preset safety margin.

[0061] Step S14: After confirming in Step S13 that the safety margin is met, first control the X-axis guide rail drive module and the Y-axis guide rail drive module to slightly translate the gripper from directly above the nominal physical center coordinate to directly above the dynamic alignment center coordinate. After completing the translational alignment, start the Z-axis actuator's descent action. The Z-axis actuator drives the gripper to begin descending at a constant speed in the vertical direction, and the fiber optic pigtail connector gradually enters the dense cable area above the target fiber optic slot. Throughout the continuous descent of the Z-axis actuator, the six-dimensional force sensor installed at the root of the gripper synchronously detects the six components of the force on the gripper at a fixed sampling period. Step S14: Extract the force component in the X-axis direction as the lateral force component in the X-axis direction, and extract the force component in the Y-axis direction as the lateral force component in the Y-axis direction. The mechanism for the occurrence of the lateral force components in the X-axis and Y-axis directions is as follows: In steps S12 and S13, the irregular polygonal outline of the remaining hole obtained by the AI ​​vision-assisted system from the top-down view only reflects the two-dimensional occlusion projection on the plane of the target fiber optic slot opening. However, the bending posture of the surrounding cables at different height levels is not consistent vertically. At one height level, the cable may intrude into the channel from the left, while at another height level, the same cable or another cable may intrude from the right front. This three-dimensional occlusion distribution that varies along the depth direction cannot be fully predicted by a single top-down view. Therefore, even if the fiber optic pigtail connector is aligned with the center coordinates of the largest inscribed rectangle on the top-down plane, as the descent depth increases, the fiber optic pigtail connector will still come into contact with cable segments bending and intruding from various directions at different depths, and will be subjected to lateral compressive forces whose direction and magnitude vary with depth. The system combines the lateral force components detected at each sampling moment along the X-axis and Y-axis into a single lateral force vector located in the XY horizontal plane using planar vector synthesis. It then calculates the resultant force amplitude based on the Pythagorean theorem and the direction angle based on the arctangent function. The resultant force amplitude represents the total lateral compression intensity experienced by the fiber optic pigtail connector at the current depth, and the direction angle represents the origin of the compression force. The real-time force detection in step S14 compensates for the information gap in step S12, where the top-view image only reflects two-dimensional projection occlusion and cannot predict the cable compression conditions at different three-dimensional heights. This enables the system to perceive obstacle distribution layer by layer along the depth direction during descent. If step S14 is missing, step S15 will not be able to know when the fiber optic pigtail connector is subjected to lateral compression, from which direction the compression comes, and how strong it is. The gripper can only blindly descend along a fixed vertical trajectory. When it encounters lateral contact, it will either bend and damage the flexible cable due to continuous pressure, or trigger an emergency stop due to overload, interrupting the operation. Both consequences will prevent the insertion process from being completed smoothly.

[0062] Step S15: Compare the resultant force amplitude of the lateral force vector with a preset compliance threshold. When the resultant force amplitude of the lateral force vector exceeds the preset compliance threshold, control the gripper of the robotic arm to generate a horizontal translation component along the normal direction of the lateral force vector. Record the displacement generated by the Z-axis actuator continuously descending in the vertical direction as the Z-axis descent component. The horizontal translation component and the Z-axis descent component are superimposed to form an inverted conical spiral descent trajectory.

[0063] Step S15 executes flexible sliding trajectory generation control based on the continuous output of the lateral force vector in step S14. The system continuously compares the resultant force amplitude of the lateral force vector calculated at each sampling moment with a preset compliance threshold. The physical meaning of the preset compliance threshold is the critical force value that distinguishes between the "elastic light contact" and "rigid compression" contact states between the fiber optic pigtail connector and the surrounding cables. When the outer wall of the fiber optic pigtail connector only makes slight contact with the bent pigtail cable sheath, the contact area is small and the elastic deformation of the cable itself is sufficient to absorb the contact force without causing a measurable offset to the axis of the fiber optic pigtail connector. The resultant force amplitude in this contact state is at a low level, and no horizontal avoidance is required. When the outer wall of the fiber optic pigtail connector makes large-area compression contact with the cable bundle, after the elastic deformation margin of the cable is exhausted, the contact force is transformed into a rigid constraint force applied to the side of the fiber optic pigtail connector, and the resultant force amplitude rises sharply. If horizontal avoidance is not performed in time, the fiber optic pigtail connector will slip uncontrollably along the direction of the compression force. The preset compliance threshold is set within the transition range between the two contact states. It is determined by repeatedly performing vertical descent tests on the fiber optic pigtail connector under simulated cable densities. The corresponding data between the resultant force amplitude and the lateral slip of the fiber optic pigtail connector are recorded in each test. A resultant force amplitude-slip curve is plotted, and the resultant force amplitude corresponding to the point where the slip starts to increase continuously from near zero is selected as the preset compliance threshold. When the resultant force amplitude of the lateral force vector is less than or equal to the preset compliance threshold, the Z-axis actuator continues to descend at a constant speed along the current vertical direction, while the X-axis and Y-axis guides remain locked. When the resultant force amplitude of the lateral force vector exceeds the preset compliance threshold, the system, while maintaining the Z-axis actuator's descent, controls the X-axis and Y-axis guide drive modules to generate a small horizontal translation component in the gripper along the normal direction of the lateral force vector in the horizontal plane. The normal direction of the lateral force vector is the direction perpendicular to the lateral force vector in the horizontal plane. There are two normal directions in the horizontal plane. The system selects the direction that makes the fiber optic pigtail connector deflect towards the side with more open space inside the irregular remaining hole polygonal contour. The specific determination method is to calculate the distance margin from the current position of the fiber optic pigtail connector to the boundary of the irregular remaining hole polygonal contour in the two normal directions respectively, and select the direction with the larger distance margin.The mechanical basis for choosing to translate along the normal direction instead of retreating directly in the opposite direction of the lateral force vector is as follows: Although retreating in the opposite direction can immediately eliminate the current contact force, the angle between the retreating direction and the target fiber optic slot axis is uncontrollable. The fiber optic pigtail connector continuously deviates from the target axis during the retreating process, and there may be another set of cables forming a new obstacle in the retreating direction, resulting in repeated oscillations of "taking a step back and encountering a new obstacle, then being forced to take another step back" and never being able to advance downwards. Translation along the normal direction allows the fiber optic pigtail connector to neither directly confront the lateral extrusion force nor retreat in a direction away from the target, but to laterally bypass the obstacle along the tangent direction of the equipotential surface of the extrusion force, avoiding the lateral obstruction at the current depth with the smallest horizontal offset, and keeping the fiber optic pigtail connector generally tending towards the target fiber optic slot axis while maintaining continuous downward advancement along the Z-axis.

[0064] The magnitude of the horizontal translation component is linearly proportional to the amount by which the resultant force amplitude of the lateral force vector exceeds the preset compliance threshold. The compliance translation gain coefficient is obtained through calibration tests and stored in the system parameter table. The physical meaning of the compliance translation gain coefficient is the horizontal translation displacement corresponding to each unit of excess force, with dimensions in millimeters per Newton. The specific calibration method for the compliance translation gain coefficient is as follows: In a standard test cabinet equipped with a simulated cable bundle, the control gripper holds the fiber optic pigtail connector and performs a vertical descent operation from directly above the target fiber slot. Multiple descent tests are performed with different candidate values ​​of the compliance translation gain coefficient. In each test, the cumulative horizontal offset of the fiber optic pigtail connector relative to the dynamic alignment center coordinates when it reaches the guide port, and whether there is any jamming between the fiber optic pigtail connector and surrounding cables during the descent, are recorded. The candidate value of the compliance translation gain coefficient that minimizes the cumulative horizontal offset and does not cause jamming throughout the descent is selected as the final calibration result. For example, the compliant translation gain coefficient is approximately 0.02 mm / Newton to 0.08 mm / Newton. A larger excess indicates stronger lateral compression and requires a larger lateral clearance, while a smaller excess results in a smaller lateral clearance. By controlling the translation amount with a linear proportional relationship rather than a fixed step size, the horizontal translation of the gripper is continuously and gradually adjusted according to the actual magnitude of the lateral force, avoiding the dilemma of overshoot due to excessive step size or insufficient clearance due to insufficient step size in fixed step size control. Under the cumulative effect of the Z-axis descent component and the horizontal translation component superimposed at each sampling moment, the actual motion trajectory of the gripper in three-dimensional space forms an inverted conical spiral descent trajectory: in the upper part of the dense cable area, due to the large range of cable bending intrusion, the horizontal translation component accumulates more, and the projection of the trajectory on the horizontal plane is a spiral or broken line shape with a large radius; as the depth increases and gradually approaches the rigid guide opening of the target fiber optic slot, the lateral intrusion of the surrounding cables is gradually replaced and constrained by the metal structure wall of the target fiber optic slot itself, the lateral squeezing force weakens, the horizontal translation component decreases, and the projection radius of the trajectory on the horizontal plane shrinks, forming an inverted conical envelope that is wider at the top and narrower at the bottom. The lateral force vector provided in step S14 is the driving signal source for the flexible sliding control in step S15. The resultant force amplitude and direction angle output at each sampling moment in step S14 directly determine whether horizontal translation occurs in step S15 at that moment, as well as the direction and magnitude of the translation. The two constitute a real-time control loop that completes the "force perception - motion response" closed loop in each sampling period. Step S15 changes the force feedback from the stop alarm trigger in the traditional scheme to a continuous motion guidance signal. When encountering lateral compression, the Z-axis descent process is not interrupted, but the fiber optic pigtail is traversed by horizontal translation, so that the fiber optic pigtail can adaptively pass through dense cable areas during continuous descent.

[0065] Step S16: Continue descending along the inverted conical spiral trajectory until the bottom of the fiber optic pigtail connector enters the guide port of the target fiber optic slot. When the six-dimensional force sensor detects a rigid wall constraint resistance in the Z-axis direction and the resultant force amplitude of the lateral force vector drops to near zero, it is determined that the fiber optic pigtail connector has entered the guide port of the target fiber optic slot.

[0066] Step S16 determines the termination condition of the inverted conical spiral descent process. As the fiber optic pigtail connector continues to descend along the inverted conical spiral trajectory, its bottom surface will eventually reach and enter the rigid guide port of the target fiber slot. The rigid guide port is made of metal or rigid engineering plastic, and its inner wall is a standard rectangular cross-section channel that matches the cross-sectional shape of the fiber optic pigtail connector. When the bottom surface of the fiber optic pigtail connector transitions from the area of ​​dense flexible cables to the interior of the rigid guide port, two synchronous changes occur in the mechanical characteristics detected by the six-dimensional force sensor: In the Z-axis direction, during the descent phase in the area of ​​dense cables, the force in the Z-axis direction mainly comes from the low-amplitude, slowly varying resistance generated by the elastic contact of the flexible cables. When the bottom surface of the fiber optic pigtail connector contacts the inlet ramp of the rigid guide port and begins to slide along the inner wall of the guide port, the change in the insertion resistance in the Z-axis direction within a single sampling period exceeds the preset rigid contact force change threshold, indicating the establishment of rigid wall constraint. The method for determining the preset rigid contact force variation threshold is as follows: During the calibration phase, multiple sets of descent tests are performed on both the flexible cable contact and the rigid guide port contact. The maximum value of the change in insertion resistance in the Z-axis direction under the flexible contact state within a single sampling period is recorded as the upper limit of the flexible contact variation. The minimum value of the change in insertion resistance in the Z-axis direction under the rigid contact state within a single sampling period is recorded as the lower limit of the rigid contact variation. The arithmetic mean of the upper limit of the flexible contact variation and the lower limit of the rigid contact variation is taken as the preset rigid contact force variation threshold. In the XY horizontal plane direction, since the inner wall of the rigid guide port is a smooth standard cross-section channel, once the fiber optic pigtail connector enters the guide port, it is no longer subjected to lateral compression from the flexible cable. The resultant force amplitude of the lateral force vector drops from the fluctuation value during the previous descent process to near zero. Near zero means that the fluctuation value drops to less than the preset lower limit threshold of the lateral force. The preset lower limit threshold for lateral force is set as the statistical upper limit of the fluctuation of the baseline resultant force amplitude of the six-dimensional force sensor caused by electrical noise and mechanical vibration under no external force contact. This statistical upper limit is obtained by suspending the gripper in an open area without any external contact during the calibration phase, continuously recording the resultant force amplitude sequence of the X-axis and Y-axis force components of the six-dimensional force sensor within a preset calibration period, and taking the statistical maximum value of the sequence. For example, the preset lower limit threshold for lateral force is approximately 0.05 Newtons to 0.15 Newtons. When the resultant force amplitude drops below the preset lower limit threshold for lateral force, it indicates that the side of the fiber optic pigtail connector has detached from the squeezing contact with the surrounding flexible cable, and the detected residual resultant force amplitude is only the sensor baseline noise level. The system's insertion resistance in the Z-axis direction exceeding the preset rigid contact force change threshold and the resultant force amplitude of the lateral force vector falling close to zero are both conditions that are met simultaneously as the basis for determining whether the fiber optic pigtail connector enters the target fiber optic slot guide port.The use of a dual-condition joint judgment instead of a single-condition judgment aims to eliminate a potentially confusing misjudgment scenario: when the bottom of the fiber optic pigtail connector is temporarily unable to descend due to being supported by a thick, laterally bent cable, the single-sampling-cycle change in the insertion resistance in the Z-axis direction may also exceed the preset threshold for the change in rigid contact force. However, the resultant force amplitude of the lateral force vector will not drop to near zero at this time because the lateral cable supporting the fiber optic pigtail connector provides support in the Z-axis direction while also applying a lateral component force in the XY plane. Since the two conditions cannot be met simultaneously, the system will not misjudge this situation as having entered the guide port. After step S16 determines that the fiber optic pigtail connector has entered the guide port of the target fiber slot, the system locks the current positions of the X-axis and Y-axis guide rails, terminates the horizontal translation control, and hands over control to step S20 to continue executing the latching and locking advancement and bottoming verification inside the guide port.

[0067] Step S10 decomposes the unstructured interference of dense cable clusters blocking and squeezing the target fiber optic slot entrance channel into two levels: static geometric reconstruction at the visual level and dynamic contact resolution at the force level, and processes them in stages according to time sequence. The visual level is completed in steps S11 to S13. Through semantic segmentation, the actual passable area contour after obstruction is obtained, and the widest channel that fits the cross-sectional shape of the fiber optic pigtail connector is found within the irregular contour by solving the maximum inscribed rectangle method. The center coordinates are extracted as the dynamic alignment position, and the starting position of the fiber optic pigtail connector is shifted from the fixed nominal physical center to the position with the best geometric safety under the current obstruction conditions, so that the initial alignment deviation is eliminated before the lowering starts. The force level is completed in steps S14 to S16. By monitoring the lateral force vector in real time and converting it into a horizontal translation along the normal direction, the fiber optic pigtail connector can adaptively avoid random lateral squeezing forces applied in different directions at each descent depth, without relying on prior modeling or pre-sorting operations regarding the three-dimensional spatial distribution of the cables. Visual and force information complement and synergize within step S10. Vision provides global initial path planning based on two-dimensional top-down observation, while force fills the gaps in local real-time obstacle avoidance capabilities in the three-dimensional depth direction that vision cannot cover at each depth along the path. The two are sequentially connected and expand from two-dimensional static to three-dimensional dynamic in terms of information dimension. If there is only a visual level and no force level, when the fiber optic pigtail descends vertically through the dynamic alignment center coordinates determined in step S13, cable protrusions encountered at different heights that are not visible from the top-down view will not be perceived or avoided, and lateral collisions will still occur. If there is only a force level and no visual level, the starting position of the fiber optic pigtail will remain at the nominal physical center coordinates that have deviated from the center of the actual passable area. Force compliance control needs to make a large horizontal translation in the initial stage of descent to avoid obstructing cables. Excessive horizontal translation will cause the fiber optic pigtail to deviate from the target fiber slot axis by a distance exceeding the coverage of the force regression capability. When the two work together, vision compresses the initial value of horizontal offset to a minimum, and force sensing only needs to perform a small amount of translation within a small range to resolve collisions along the way. The cumulative amount of horizontal translation is much smaller than the amount of translation when force sensing is used alone. The deviation between the final slot position of the fiber optic pigtail connector and the axis of the target fiber optic slot is constrained within the self-correction tolerance range of the guide port. The resistance signal reflected by the force-displacement curve in step S20 has physical interpretability because the coaxiality of the fiber optic pigtail connector and the slot channel is guaranteed, enabling subsequent radius of curvature analysis and side projection depth inversion to run on reliable mechanical input.

[0068] Step S20: After the fiber optic pigtail connector enters the guide port of the target fiber slot, a force-displacement curve is plotted. When a resistance peak appears on the force-displacement curve, the radius of curvature of the curve in the vicinity of the resistance peak is calculated. The radius of curvature is compared with a preset radius of curvature threshold to determine whether the bottoming is true or false. When it is determined to be a false bottoming, the side projection chord length is obtained. The under-insertion depth is inverted based on the side projection chord length. High-frequency micro-oscillation compensation insertion is performed according to the under-insertion depth until the force-displacement curve shows a true locking feature.

[0069] Specifically, step S20 is executed after step S16 determines that the fiber optic pigtail connector has entered the guide port of the target fiber optic slot. Step S16 determines that the connector has entered the slot based on the fact that the change in the insertion resistance in the Z-axis direction within a single sampling period exceeds a preset threshold for the change in rigid contact force, and the resultant force amplitude of the lateral force vector drops to near zero. At this point, the bottom surface of the fiber optic pigtail connector has transitioned from the messy cable area above to the rigid metal or hard engineering plastic channel inside the target fiber optic slot. Subsequent Z-axis downward pressing will cause the fiber optic pigtail connector to continue to penetrate deeper into the channel inside the slot until the locking mechanism is triggered. In a brand new or well-maintained fiber optic slot, the elastic spring is pushed open by the protrusion on the side of the fiber optic pigtail connector and then suddenly springs back to lock at the moment of locking. The insertion resistance in the Z-axis direction experiences a sudden drop from its peak value to near zero within a very short displacement range. Based on this, the system determines that the insertion is complete and performs force recovery. However, in racks that have been in use for many years, the elastic springs inside some fiber optic slots experience metal fatigue and elastic decay due to long-term, high-frequency insertion and removal operations. Alternatively, fiber debris and dust suspended in the air inside the rack gradually accumulate between the springs and the fiber optic pigtail connector, forming a dry friction pair. When the fiber optic pigtail connector is pressed down to the half-depth position in these aged slots, the dry friction force generated by the deformed springs or foreign particles continuously increases with the insertion depth. After reaching the peak static friction force, it slowly decreases due to the sliding friction coefficient being lower than the static friction coefficient, forming a curve of resistance that first rises and then falls. This curve closely resembles the resistance that first rises and then falls during actual latching. If the system only uses the single criterion of "resistance starting to decrease after exceeding the peak" to perform force recovery, it will misjudge the false force peak generated by dry friction as the insertion being complete and prematurely terminate the pressing, leaving the fiber optic pigtail connector actually in a half-insertion state at the half-depth position. Meanwhile, the gripper itself is a rectangular block structure, which completely blocks the area directly below when vertically pressed in. The optical axis of the AI ​​vision-assisted system's overhead camera is cut off by the gripper body, making it impossible to penetrate the gripper to observe the actual engagement depth between the fiber optic pigtail connector and the target fiber optic slot. The vertical visual verification channel is physically blocked. Step S20, through the progressive engagement from steps S21 to S25, uses the radius of curvature of the curve in the neighborhood of the extreme point of the force-displacement curve as a geometric criterion to distinguish between real snap-locking and dry friction artifacts. A micro-vision module installed on the side and rear of the gripper acquires the side projection chord length of the fiber optic pigtail connector exposed outside the slot from an oblique angle and inverts the under-insertion depth. High-frequency micro-oscillation transforms the static friction under dry friction into dynamic friction, thereby flexibly advancing the fiber optic pigtail connector until real snap-locking. This method forms a closed-loop verification and compensation insertion process in three dimensions: force analysis, visual inversion, and oscillatory advancement.

[0070] Further, step S20 includes:

[0071] Step S21, see Figure 5 After the fiber optic pigtail connector enters the guide port of the target fiber optic slot, the Z-axis actuator is controlled to continue pressing down at a constant speed in the vertical direction. During the continuous pressing down of the Z-axis actuator, the displacement encoder on the Z-axis actuator records the cumulative downward displacement of the Z-axis at a fixed sampling period, and at the same time, the insertion resistance in the Z-axis direction is recorded by a six-dimensional force sensor at the same sampling period. The force-displacement curve is plotted point by point with the cumulative downward displacement of the Z-axis as the abscissa and the insertion resistance in the Z-axis direction as the ordinate. The force-displacement curve is continuously monitored to detect the peak point of the resistance.

[0072] Step S21: After confirming in Step S16 that the fiber optic pigtail connector has entered the guide port, control the Z-axis actuator to continue pressing down at a constant speed in the vertical direction, pushing the fiber optic pigtail connector deeper into the target fiber slot's internal channel, and immediately start data acquisition synchronously. The displacement encoder integrated on the Z-axis actuator is an incremental photoelectric encoder, which outputs the cumulative downward displacement of the Z-axis actuator relative to its position at the time of entering the guide port in each sampling cycle. The displacement encoder has a resolution on the sub-micrometer scale, capable of capturing minute displacement increments during high-frequency micro-oscillations. A six-dimensional force sensor synchronously records the Z-axis insertion resistance experienced by the gripper in the Z-axis direction with the same fixed sampling cycle as the displacement encoder. Using the cumulative downward displacement of the Z-axis acquired at each sampling moment as the abscissa and the Z-axis insertion resistance as the ordinate, a continuous force-displacement curve is plotted point by point in memory. The cumulative descent displacement along the Z-axis, rather than time, is used as the abscissa because the descent speed of the Z-axis actuator during its advancement within the slot is not constant. Before contacting the latch, the descent speed is relatively stable due to lower channel resistance. As the resistance increases near the latch, the speed gradually decreases due to resistance feedback. During the high-frequency micro-oscillation phase in step S24, the speed exhibits a rapid, sinusoidal alternation. Using time as the abscissa, the difference in time span corresponding to the same physical displacement at different speeds will cause the force-displacement curve to be non-uniformly stretched or compressed along the time axis. The same physical and mechanical event will exhibit different curve geometries under different speed conditions. The radius of curvature calculation performed on the curve shape in step S22 will drift due to speed changes, losing comparability and consistency across different operating conditions. With the cumulative descent displacement along the Z-axis as the abscissa, the geometry of the force-displacement curve is determined solely by the resistance encountered by the fiber optic pigtail connector at a certain insertion depth, independent of the instantaneous velocity of the Z-axis actuator upon reaching that depth. This ensures that the radius of curvature calculation in step S22 remains stable and repeatable under different descent speeds, providing a geometric analysis basis for determining the authenticity of bottoming out that is independent of fluctuations in motion parameters. Without the continuous plotting of the force-displacement curve in step S21, step S22 would be unable to obtain continuous data segments in the neighborhood of the resistance peak point for radius of curvature calculation. Furthermore, the continuous monitoring of whether new sharp peaks appear during the high-frequency micro-oscillation phase in step S25 would be impossible due to the lack of a real-time updated force-displacement curve, and the entire bottoming-out determination link would be broken due to the absence of an input data source.

[0073] Step S22, see below. Figure 5When the insertion resistance in the Z-axis direction on the force-displacement curve reaches a local maximum and begins to decrease, the local maximum point is marked as the resistance peak point. Curve data segments within a preset window width range before and after the resistance peak point are extracted. The second derivative is calculated on the curve data segments to obtain the curve curvature radius of the neighborhood of the resistance peak point. The curve curvature radius is compared with a preset curvature radius threshold. When the curve curvature radius is less than the preset curvature radius threshold, the resistance peak point is determined to correspond to a real latching event, and force recovery and release of the gripper are executed to complete the insertion operation. When the curve curvature radius is greater than or equal to the preset curvature radius threshold, the resistance peak point is determined to correspond to a false bottoming event, and the downward pressing action of the Z-axis actuator is frozen. The current cumulative downward displacement of the Z-axis is marked as the false bottoming frozen displacement.

[0074] See Figure 6 This is a schematic diagram comparing the force-displacement curves of actual latch engagement and false bottoming provided in the embodiments of this application. Figure 6 The diagram illustrates two sets of force-displacement curves, with the cumulative Z-axis displacement as the abscissa and the Z-axis insertion resistance as the ordinate. These correspond to the actual locking event and the dry friction illusion event, respectively. The diagram also marks the resistance peaks in the curves, as well as the sharp peak characteristics and small radius of curvature associated with the actual locking event, and the gentle circular peak characteristics and large radius of curvature associated with the dry friction illusion. Step S22 performs peak detection and radius of curvature calculation on the force-displacement curves continuously plotted in step S21. The ordinate sequence of the force-displacement curves is continuously scanned using a sliding window. When the Z-axis insertion resistance monotonically increases to a local maximum value within a series of sampling points and then monotonically decreases, and the decreasing trend from the local maximum value point continues to exceed the preset number of confirmed sampling points, this local maximum value point is marked as... Figure 6The resistance peak points are marked on both sets of curves. The preset number of sampling points is set to prevent occasional sensor noise-induced jitter from being mistakenly identified as a peak and then the downward trend. The value is obtained by doubling the longest continuous number of sampling points that continuously decrease due to sensor noise during the calibration test. After marking the resistance peak point, curve data segments within preset window widths before and after the resistance peak point are extracted. The preset window width is measured in terms of the cumulative downward displacement along the Z-axis. The value is based on the typical displacement span corresponding to the rise from the resistance peak to the steady state during the standard buckle locking process. This value is obtained by statistically analyzing the force-displacement curves obtained after performing insertion tests on multiple sets of standard fiber optic slots during the factory calibration phase. For the extracted curve data segments, the curvature of the force-displacement curve in the neighborhood of the resistance peak point is obtained using the numerical second derivative calculation method. The curvature at a point on the curve is defined as the degree to which the curve deviates from a straight line in the neighborhood of that point. The radius of curvature is the reciprocal of the curvature, and its physical meaning is the radius of the arc that can locally fit the curve at that point. The calculation of the numerical second derivative uses a central difference scheme. By taking several equidistant sampling points before and after the resistance peak point, the resistance value and the cumulative Z-axis displacement are inserted into the central difference formula to obtain discrete approximate values ​​of the first and second derivatives. Then, combined with the curvature formula (curvature equal to the absolute value of the second derivative divided by the square of the first derivative plus 1 raised to the power of 1.5), the curvature value is calculated, and its reciprocal is the radius of curvature of the curve. The calculation step size of the central difference scheme is equal to the displacement increment between two consecutive sampling points of the displacement encoder. This step size is determined by the displacement encoder resolution and sampling period and does not require additional setting.

[0075] The physical principle distinguishing genuine snap-locking from the illusion of dry friction is as follows: During rigid snap-locking, the elastic spring is linearly compressed when pushed open by the protrusion on the side of the fiber optic pigtail connector. The insertion resistance in the Z-axis direction increases approximately linearly with displacement. At the instant the protrusion just passes the spring's locking point, the elastic restoring force of the spring suddenly reverses from hindering insertion to assisting locking. The insertion resistance in the Z-axis direction drops sharply from its peak within a displacement range not exceeding a fraction of a millimeter. The force-displacement curve forms a sharp, approximately zigzag angle at the resistance peak point. The curvature at this angle is extremely high, and the radius of curvature of the curve is extremely small. Figure 6 The sharp peak characteristic of the actual snap-locking curve in the middle is characterized by a small radius of curvature; during the dry friction transition process, the transition of friction between the deformed spring or foreign particles and the fiber optic pigtail connector from static friction to sliding friction is a gradual process accompanied by the gradual slippage of the micro-contact surface. The rate of decline of the insertion resistance in the Z-axis direction from the peak value is modulated by the continuous change in the surface roughness and contact area of ​​the friction pair material. The force-displacement curve shows a gentle arc-shaped transition at the resistance peak point, with a small curvature and a large radius of curvature. Figure 6The pseudo-dry friction curve exhibits a gentle, rounded peak with a large radius of curvature. The calculated radius of curvature is compared with a preset radius of curvature threshold. The preset radius of curvature threshold is determined as follows: During factory calibration, multiple insertion tests are performed on both brand-new standard fiber optic slots and fiber optic slots that have undergone accelerated aging. For the brand-new slot group, the maximum radius of curvature during each actual locking is recorded as the upper limit of the actual locking radius of curvature. For the aged slot group, the minimum radius of curvature during each pseudo-peak of dry friction is recorded as the lower limit of the pseudo-bottoming radius of curvature. The preset radius of curvature threshold is the arithmetic mean of the upper limit of the actual locking radius of curvature and the lower limit of the pseudo-bottoming radius of curvature, ensuring that the threshold falls in the middle of the interval between the radius of curvature distributions of the two types of events, thus distinguishing between them. For example, when the upper limit of the actual locking radius of curvature is approximately 0.3 mm and the lower limit of the pseudo-bottoming radius of curvature is approximately 1.2 mm in the calibration test, the preset radius of curvature threshold is approximately 0.75 mm.

[0076] When the radius of curvature of the curve is less than the preset radius of curvature threshold, the curve trend at the resistance peak point is as follows: Figure 6 The sharp peak characteristic corresponding to a true snap-lock event is identified as a genuine snap-lock event. The Z-axis actuator immediately stops pressing down and slightly lifts up a preset springback distance to release residual pressure. This prevents continued pressure after the snap-lock is engaged, which could cause the latch to bear a continuous load exceeding its compressive strength, resulting in plastic deformation or fracture. The preset springback distance must simultaneously meet two constraints: first, it must be greater than or equal to the elastic recovery distance corresponding to the residual compression deformation of the elastic spring after snap-locking, to ensure that the residual pressure is completely released; second, it must be less than the effective locking stroke of the snap-lock, to avoid pulling the engaged snap-lock out by exceeding the locking stroke. The specific value of the preset springback distance is obtained by performing multiple snap-locking tests on the standard fiber optic slot during the calibration phase, gradually lifting the device, and recording the minimum lifting distance that causes the residual Z-axis insertion resistance to fall back to the sensor baseline noise level. For example, the preset springback distance is approximately 0.1 mm to 0.3 mm. The jaws are then released to complete the insertion operation. When the curve radius of curvature is greater than or equal to a preset radius of curvature threshold, the curve trend at the resistance peak point is as follows: Figure 6The smooth arc peak characteristic corresponding to the dry friction illusion is determined to be a false bottoming event. The downward pressing action of the Z-axis actuator is immediately frozen, the current Z-axis position is locked, the cumulative downward displacement of the Z-axis at this time is recorded and marked as the false bottoming frozen displacement, and then the process proceeds to step S23. The radius of curvature of the curve is used as the criterion for judgment, rather than the absolute magnitude of the resistance peak or the slope of the resistance descent. This is because different models of fiber optic slots have different elastic stiffness coefficients of the snap-fit ​​springs, and the absolute magnitude of the resistance peak during actual locking can vary by several times between different models. Using a fixed force threshold to distinguish between genuine and counterfeit locking will frequently result in misjudgments due to model differences. Although the slope of the resistance descent can also reflect the degree of descent, it is affected by the descent speed of the Z-axis actuator. The descent of force in the same displacement domain corresponds to different slope values ​​in the time domain at different speeds. Judging by the slope in the force-time domain requires an additional speed normalization step. However, the radius of curvature of the curve, as a pure force-displacement geometric quantity, naturally eliminates the interference of speed and model differences, so that the judgment result depends only on the sharpness of the mechanical change at the resistance peak point, which is an intrinsic feature directly corresponding to genuine and counterfeit snap-fit ​​locking. The design of selecting the cumulative downward displacement of the Z-axis as the horizontal axis in step S21 ensures that the curvature radius calculation result in step S22 is not affected by the speed fluctuation of the Z-axis actuator. If step S21 uses time as the horizontal axis, the curvature radius calculated in step S22 will change with the pressing speed. The same real latching event will produce different curvature radius values ​​under different pressing speeds, and the preset curvature radius threshold will not be applicable to multiple speed conditions at the same time.

[0077] Step S23: In the false bottoming-out frozen state, the macro vision module is called to acquire the side image of the fiber optic pigtail connector from a preset oblique viewing angle. The trapezoidal projection contour of the fiber optic pigtail connector exposed outside the target fiber slot is identified from the side image. Edge detection and line fitting are performed on the trapezoidal projection contour. The side pixel length extending along the height direction of the fiber optic pigtail connector in the trapezoidal projection contour is extracted and recorded as the side projection chord length.

[0078] Step S23: Immediately after step S22 determines a false bottoming out and freezes the Z-axis actuator, side vision acquisition is initiated. In this state, the AI ​​vision-assisted system's top-view camera is unable to observe the engagement status of the fiber optic pigtail connector and the target fiber optic slot due to the gripper's obstruction. The macro vision module is installed to the side and rear of the gripper, with its optical axis at a non-zero preset oblique angle to the Z-axis. This bypasses the vertical obstruction of the gripper body, allowing observation from the oblique side of the section of the fiber optic pigtail connector exposed above the target fiber optic slot opening. Determining the preset oblique angle requires considering two constraints: if the angle is too small, the optical axis of the macro vision module is close to the vertical direction and may still be obstructed by the gripper edge; furthermore, the projection length of the fiber optic pigtail connector's side in the image is too short due to perspective projection, hindering accurate extraction. If the angle is too large, the optical axis of the macro vision module is close to the horizontal direction, and the target fiber optic slot opening appears as an extremely narrow line in the field of view, making it impossible to identify the boundary between the fiber optic pigtail connector and the slot opening. During the factory installation phase, by gradually adjusting the installation angle, acquiring side images at different known insertion depths, and evaluating the recognizability of the side projection of the fiber optic pigtail connector in the images, an angle value that satisfies both the sensitivity to changes in the side projection length and the recognizability of the slot opening is selected as the preset oblique viewing angle. For example, the preset oblique viewing angle is approximately between 30° and 45°.

[0079] The macro vision module acquires a side image under a pseudo-bottom-freeze state. The fiber optic pigtail connector itself is a regular cuboid shape. The portion of it exposed above the target fiber slot appears as an approximately trapezoidal projection outline due to perspective projection effect at the oblique viewing angle of the macro vision module. The bottom edge of the fiber optic pigtail connector closer to the macro vision module projects a longer image in the image, while the top edge farther from the macro vision module projects a shorter image. The two horizontal edges are connected by two side edges extending along the height direction, forming a trapezoid that is narrower at the top and wider at the bottom. Canny edge detection is performed on the side image to extract the set of edge pixels. Hough line detection is performed on the subset of pixels belonging to the trapezoidal projection outline to fit the lines containing the four sides of the trapezoid. From the four lines, two side lines extending along the height direction of the fiber optic pigtail connector are identified. The more complete and clearer side line in the image is selected, and its pixel length in the image is measured and recorded as the side projection chord length. Both Canny edge detection and Hough line detection are well-known algorithms in the field of image processing. The former extracts edge pixels at gray-level step positions in the image through gradient calculation and non-maximum suppression, while the latter votes edge pixels in the parameter space to detect sets of pixels with collinearity and fits a straight line equation. The physical meaning of the side projection chord length is the perspective projection length of the actual height of the fiber optic pigtail connector exposed above the target fiber slot opening on the imaging surface of the macro vision module: the larger the exposed height, the larger the side projection chord length, and the smaller the exposed height, the smaller the side projection chord length, with a monotonically decreasing correspondence between the two. Step S23 transforms the vertical visual blind zone problem into a geometric length measurement problem of the exposed part of the fiber optic pigtail connector observed from the oblique side. Without adding an additional ranging sensor or changing the outer contour structure of the gripper, an intermediate quantity reflecting the insertion depth state is obtained only through the installed macro vision module and image geometric analysis, providing a quantifiable visual input for the depth inversion in step S24. If the side visual acquisition in step S23 is missing, the side projection chord length data cannot be obtained in step S24, the inversion of the under-insertion depth will not be able to be executed due to the lack of input, the propulsion target of the high-frequency micro-oscillation in step S24 will not be able to be quantitatively set, and the system will lose the ability to judge how far the fiber optic pigtail connector is from the actual locking position in the false bottoming state.

[0080] Step S24: Input the side projection chord length into the pre-constructed insertion depth mapping model to obtain the under-insertion depth of the fiber optic pigtail connector from the actual latching position; Based on the under-insertion depth, control the Z-axis actuator to perform high-frequency micro-oscillation with a preset micro-oscillation amplitude and preset micro-oscillation frequency on the basis of the false bottoming-freezing displacement. During the high-frequency micro-oscillation, continuously call the micro-vision module to re-acquire the side image and recalculate the under-insertion depth, while continuously monitoring whether a new resistance peak point appears on the force-displacement curve;

[0081] See Figure 7 The method for constructing the insertion depth mapping model includes: taking a standard fiber optic slot and a standard fiber optic pigtail connector, controlling the Z-axis actuator to gradually insert the standard fiber optic pigtail connector into the standard fiber optic slot at fixed step lengths; at each step length position, calling the macro vision module to acquire a side image of the standard fiber optic pigtail connector from a preset oblique viewing angle, extracting the side projection chord length from the side image, and simultaneously recording the known remaining depth value of the Z-axis actuator from the actual locking position; establishing a one-to-one mapping relationship table between the side projection chord length and the known remaining depth value, and storing the mapping relationship table as an insertion depth mapping model.

[0082] Step S24 receives the side projection chord length output from step S23, inverts the side projection chord length into the under-insertion depth using a pre-built insertion depth mapping model, and then initiates high-frequency micro-oscillation for compensation insertion. The insertion depth mapping model is constructed during the system factory calibration phase according to the following process: Take a set of standard fiber optic slots and standard fiber optic pigtail connectors of the same model as those used in actual operation, install the standard fiber optic pigtail connectors on the grippers, and control the Z-axis actuator to gradually advance downwards from the starting position of the standard fiber optic slot opening at a fixed step size. For example, the fixed step size is approximately 0.1 mm. At each step size position, pause the descent of the Z-axis actuator, call the macro vision module to acquire the side image of the standard fiber optic pigtail connector from the same preset oblique viewing angle as in actual operation, extract the side projection chord length at that step size position according to the same Canny edge detection and Hough line detection process as in step S23, and simultaneously calculate the known remaining depth value of the Z-axis actuator from the actual locking position at that step size position based on the difference between the current displacement of the Z-axis actuator and the pre-measured and confirmed actual locking position of the standard fiber optic slot. The actual locking position was determined before calibration by manually inserting the standard fiber optic pigtail connector fully into the standard fiber optic slot until the latch was engaged, and then reading the displacement encoder reading of the Z-axis actuator. (See also...) Figure 7 After traversing all step length positions, a mapping table is established with the side projection chord length as the input column and the known remaining depth value as the output column. This mapping table is stored as an insertion depth mapping model. If not all step lengths have been traversed, the Z-axis actuator continues to be controlled to insert the standard fiber optic pigtail connector step by step at a fixed step length. For example, part of the data in the mapping table is shown in Table 1.

[0083] Table 1. Examples of mapping relationships

[0084] Side projection chord length (pixels) The remaining depth (in millimeters) is known. 186 3 158 2.5 131 2 103 1.5 76 1 48 0.5 21 0.1

[0085] After step S23 outputs the side projection chord length in the current pseudo-bottoming state, the system inputs it into the mapping table to perform table lookup and linear interpolation: find the two adjacent rows in the table where the side projection chord length falls, calculate the corresponding remaining depth value using linear interpolation, and output this value as the under-insertion depth of the current fiber optic pigtail connector from the actual latching position. The under-insertion depth is obtained by using a side projection chord length lookup table instead of actively measuring using laser ranging during insertion. This is because in dense cabinet environments, the mating gap between the fiber optic pigtail connector and the target fiber optic slot is only a few tenths of a millimeter and is completely surrounded by the grippers. The laser ranging transmitter cannot obtain a direct path into the slot within the constraints of the gripper's outer dimensions, making active measurement physically infeasible. The macro vision module, observing the exposed portion of the fiber optic pigtail connector from an oblique side, is a passive vision measurement method. It does not require transmitting a probe signal into the slot; it can be completed using only the natural reflected light from the outer surface of the fiber optic pigtail connector under ambient lighting conditions, making it feasible under spatial constraints.

[0086] After obtaining the under-insertion depth, the system controls the Z-axis actuator to initiate high-frequency micro-oscillation based on the position of the false bottoming-out frozen displacement. The high-frequency micro-oscillation motion is that the Z-axis actuator performs sinusoidal axial reciprocating motion along the vertical direction with a preset micro-oscillation amplitude as half the amplitude and a preset micro-oscillation frequency as the reciprocating frequency. The preset micro-oscillation amplitude is limited to the range of 0.01 mm to 0.05 mm. The setting is based on the fact that the elastic travel of the standard fiber optic slot latch spring is usually on the order of 0.3 mm to 0.8 mm. The preset micro-oscillation amplitude must be much smaller than the lower limit of the elastic travel, so that even if the fiber optic pigtail connector happens to reach the vicinity of the spring latch point during a certain positive half-cycle of oscillation, the single advance distance is not enough to forcibly cross the latch point without establishing correct alignment and damage the spring. The preset micro-oscillation frequency is set based on the following principle: the frequency needs to be high enough to allow the contact interface between the fiber optic pigtail connector and the deformable spring or foreign object to undergo multiple state transitions from static to relative sliding within a very short time. Tribological principles indicate that the average effective friction coefficient of the micro-contact surface during repeated high-frequency start-stop processes is lower than the static friction coefficient in a continuously static state. The macroscopic equivalent behavior of the contact interface transforms from static friction to dynamic friction, allowing the fiber optic pigtail connector to achieve a small net downward displacement increment in each forward propulsion half-cycle because the dynamic friction resistance is lower than the static friction resistance that cannot be overcome under constant thrust mode. For example, the preset micro-oscillation frequency is approximately 50 Hz to 200 Hz. Simultaneously, the overall downward offset of the Z-axis actuator is superimposed on the sinusoidal oscillation by increasing a small downward offset in each oscillation cycle. The step size of the small downward offset does not exceed one-tenth of the preset micro-oscillation amplitude, ensuring that the net downward propulsion speed is extremely slow. The fiber optic pigtail connector approaches the actual locking position by gradually and flexibly wedging rather than by continuous constant force pushing.

[0087] During the high-frequency micro-oscillation process, the system performs two continuous monitoring operations in parallel. In the visual monitoring channel, at preset visual acquisition intervals, the system calls the macro vision module in step S23 to re-acquire a frame of the side image of the fiber optic pigtail connector. Following the same process as in step S23, the side projection chord length is re-extracted, and the re-extracted side projection chord length is input into the insertion depth mapping model in step S24 to recalculate the current under-insertion depth, monitoring whether the under-insertion depth has approached zero. Approaching zero means that the detected current under-insertion depth is less than a preset under-insertion depth convergence threshold. The preset under-insertion depth convergence threshold is the larger of the upper limit of the mapping accuracy of the insertion depth mapping model during the calibration phase and the resolution of the displacement encoder; its physical meaning is the minimum remaining under-insertion depth that the system can reliably distinguish. For example, the preset under-insertion depth convergence threshold is approximately 0.05 mm to 0.15 mm. The preset visual acquisition interval is the sum of the times of several complete oscillation cycles corresponding to the preset micro-oscillation frequency. This ensures that the Z-axis actuator is precisely near the middle equilibrium position of the oscillation during each visual acquisition, avoiding periodic fluctuations in the side projection chord length caused by the superposition of oscillation displacements due to image acquisition at extreme oscillation positions. In the force monitoring channel, the system continuously observes whether a new resistance peak point appears in the force-displacement curve updated in real time in step S21. If a new resistance peak point appears, the radius of curvature of the curve in the neighborhood of the new resistance peak point is calculated using the same second derivative method of the central difference value in step S22 and compared with a preset radius of curvature threshold. The two monitoring channels run in parallel in time and are logically ANDed in the judgment logic.

[0088] Step S25: When the under-insertion depth approaches zero and the radius of curvature of the curve in the neighborhood of the newly appeared resistance peak point is less than the preset radius of curvature threshold, it is determined that the fiber optic pigtail connector has reached the actual locking position, the high-frequency micro-oscillation is stopped, the force is recovered and the clamp is released to complete the insertion operation; if the above conditions are still not met after the high-frequency micro-oscillation duration exceeds the preset maximum oscillation duration, the fiber optic pigtail connector is raised to a safe height and an abnormal report is generated.

[0089] Step S25 sets the termination and exit conditions for the high-frequency micro-oscillation process. When the visual monitoring channel determines that the under-insertion depth is close to zero and the force monitoring channel determines that the radius of curvature of the curve in the neighborhood of the newly appearing resistance peak point is less than the preset radius of curvature threshold, the simultaneous satisfaction of both conditions indicates that the fiber optic pigtail connector has reached the actual snap-lock position in physical space and has generated the sharp peak corresponding to the actual snap-lock in mechanical characteristics. A dual-condition joint determination is adopted instead of any single condition. If only the under-insertion depth approaching zero is used as the termination condition without verifying the appearance of the force-sensing sharp peak, an extreme scenario exists: the locking spring of the target fiber optic slot has completely failed and lost its elastic restoring force; the fiber optic pigtail connector is pushed to the locking position by high-frequency micro-oscillation, but the spring cannot spring back to lock; the under-insertion depth reaches zero, but the locking is not engaged; the fiber optic pigtail connector slides out of the slot due to gravity or cable tension after the clamp is released because of the lack of locking constraint. Conversely, if only the appearance of the force-sensing sharp peak is used as the termination condition without verifying whether the under-insertion depth has reached zero, the fiber optic pigtail connector may encounter a burr or foreign object on the inner wall of the slot during high-frequency micro-oscillation, generating a mechanical abrupt sharp peak. The system might mistakenly judge this as locking and terminate prematurely before the fiber optic pigtail connector reaches the actual locking position. The dual-condition joint determination eliminates these two types of misjudgment scenarios, ensuring that the termination condition simultaneously verifies both spatial positioning and mechanical locking. When both conditions are met, the high-frequency micro-oscillation stops immediately, the Z-axis actuator stops oscillating and slightly raises the preset springback distance to release residual pressure, releases the gripper to complete the insertion operation and resets.

[0090] If the high-frequency micro-oscillation continues for longer than the preset maximum oscillation duration and both conditions are still not met simultaneously, it is determined that the target fiber optic slot has physical damage that cannot be recovered by the high-frequency micro-oscillation. The preset maximum oscillation duration is set based on the net advance rate determined by the product of the slight downward offset step size of each oscillation cycle and the preset micro-oscillation frequency. The theoretical maximum time required to advance the fiber optic pigtail connector from the false bottoming position to the actual locking position is multiplied by a preset safety factor to obtain the preset maximum oscillation duration. The preset safety factor is greater than 1 to cover adverse working conditions where the dry friction coefficient is higher than the typical value during calibration. For example, the preset safety factor is approximately 3. After the timeout, the high-frequency micro-oscillation stops, and the Z-axis actuator is controlled to raise the fiber optic pigtail connector to a safe height completely detached from the target fiber optic slot. An anomaly report is generated, recording the row and column number of the target fiber optic slot and the amount of false bottoming-freeze displacement, etc., awaiting manual intervention for inspection and handling.

[0091] Step S20 uses the radius of curvature of the force-displacement curve—a geometric quantity—to distinguish between real latching and dry friction artifacts into two categories of events with separable intervals in geometric characteristics. This makes the bottoming-out determination no longer dependent on the absolute magnitude of the resistance peak or the descent slope, two non-robust indicators affected by slot model differences and downward pressure speed fluctuations. The mechanism of inverting under-insertion depth using the side projection chord length recovers the depth information in the vertical visual blind spot obscured by the gripper through indirect measurement. It utilizes the non-coaxial mounting relationship between the macro vision module and the gripper to bypass the physical constraints of obstruction. By mapping the pixel length of the two-dimensional image to three-dimensional spatial depth using the calibrated perspective projection geometry, it expands the depth perception dimension of the system without adding additional sensors. High-frequency micro-oscillations, with amplitudes much smaller than the elastic travel of the latch and sufficiently high frequencies, repeatedly switch the contact interface from a static friction state to a dynamic friction state. This causes the dry friction barrier, which was originally impossible to overcome by constant thrust due to excessive static friction, to be gradually eliminated after the equivalent friction coefficient is reduced under high-frequency switching. The fiber optic pigtail connector is flexibly advanced with extremely small net displacement increments obtained in each oscillation cycle. The single advancement amount is far below the dangerous displacement threshold that may cause damage to the latch or slot. The concatenation of the radius of curvature determination, the side projection chord length depth inversion, and the high-frequency micro-oscillation compensation insertion forms a complete closed loop in step S20: "false detection - quantification deficiency - flexible compensation - dual-channel termination confirmation". Step S22's radius of curvature determination identifies false bottoming events in the force dimension and freezes the pressure to prevent the system from terminating at the wrong position. Steps S23 and S24's side visual inversion quantifies the distance between the frozen position and the real target position in the visual dimension. Step S24's high-frequency micro-oscillation gradually eliminates this distance in a controlled manner in the mechanical dimension. Step S25's dual-condition joint termination cross-verifies the position status in two independent channels, force and vision. The satisfaction of either channel alone is insufficient to trigger termination. Steps S20 and S10 are sequentially connected. Step S10 safely delivers the fiber optic pigtail connector from the messy cable area to the target fiber optic slot guide port, ensuring that the coaxiality of the fiber optic pigtail connector and the slot channel is within the self-correcting tolerance range of the guide port. Based on this coaxiality guarantee, step S20 begins the acquisition of force-displacement curves and curvature analysis. If the fiber optic pigtail connector deviates too far from the axis when entering the guide port, the uneven friction between it and the inner wall of the slot will be superimposed on the latching resistance, causing the force-displacement curve to produce stray peaks unrelated to the latching. The curvature radius analysis will be run on noisy data and output unreliable conclusions. Step S10 provides the coaxiality prerequisite for step S20. Under this prerequisite, step S20 completes the closed-loop verification from artifact recognition to compensated insertion to in-place confirmation. Together, they cover the two different types of perception failures—visual occlusion and force deception—in the complete operation process from messy cable passage to aging slot locking.

[0092] Example 2:

[0093] This embodiment, based on Embodiment 1, provides an end-effector pose dynamic compensation system based on the fusion of visual servoing and force feedback, such as... Figure 8 As shown, it includes:

[0094] Dynamic alignment center module: used to acquire top view images of the target fiber optic slot and surrounding area, segment out the irregular remaining hole polygonal outline after being blocked by surrounding cables from the top view image, and obtain the dynamic alignment center coordinates based on the irregular remaining hole polygonal outline.

[0095] Inverted conical spiral feeding module: Based on the dynamic alignment center coordinates, the lateral force vector is obtained. According to the lateral force vector, an inverted conical spiral descent trajectory is generated. The gripper of the robotic arm is controlled to feed the fiber optic pigtail connector into the guide port of the target fiber optic slot along the inverted conical spiral descent trajectory.

[0096] Bottom-out authenticity determination module: After the fiber optic pigtail connector enters the guide port of the target fiber slot, it draws a force-displacement curve. When the force-displacement curve shows a resistance peak point, it calculates the curve curvature radius of the neighborhood of the resistance peak point and compares the curve curvature radius with a preset curvature radius threshold to determine the authenticity of bottom-out.

[0097] High-frequency micro-oscillation compensation module: When a false bottoming out is determined, the side projection chord length is obtained, the under-insertion depth is inverted based on the side projection chord length, and high-frequency micro-oscillation compensation insertion is performed according to the under-insertion depth until the force-displacement curve shows the true locking characteristics.

[0098] Furthermore, in the dynamic alignment center module, the method for obtaining the polygonal outline of the irregular remaining holes includes:

[0099] Semantic segmentation is performed on the top view image to distinguish the passable area of ​​the target fiber optic slot opening from the obstruction area of ​​the surrounding cables. The outer boundary of the passable area is extracted and fitted into an irregular residual hole polygonal contour. The coordinates of each vertex of the irregular residual hole polygonal contour are stored.

[0100] The method for obtaining the dynamic alignment center coordinates includes:

[0101] Read the nominal length and nominal width values ​​of the cross-section of the fiber optic pigtail connector, calculate the aspect ratio of the cross-section of the fiber optic pigtail connector, solve for the maximum inscribed rectangle within the irregular remaining hole polygonal outline with the aspect ratio of the cross-section of the fiber optic pigtail connector as the constraint condition, and extract the center coordinates of the maximum inscribed rectangle as the dynamic alignment center coordinates.

[0102] Furthermore, the inverted conical spiral is fed into the module, and the method for obtaining the lateral force vector includes:

[0103] When the length of the short side of the largest inscribed rectangle meets the safety margin requirement, the robotic arm is controlled to translate to directly above the dynamic alignment center coordinates; from directly above the dynamic alignment center coordinates, the Z-axis descent is executed. During the Z-axis descent, the lateral force components in the X-axis direction and the Y-axis direction are detected in real time by a six-dimensional force sensor, and the lateral force components in the X-axis direction and the Y-axis direction are combined into a lateral force vector.

[0104] The methods and systems of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0105] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0106] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for dynamic end-effector pose compensation based on the fusion of visual servoing and force feedback, characterized in that, The method includes: A top-view image of the target fiber optic slot and its surrounding area is acquired. The polygonal outline of the irregular remaining hole after being blocked by the surrounding cables is segmented from the top-view image. The dynamic alignment center coordinates are obtained based on the polygonal outline of the irregular remaining hole. The lateral force vector is obtained based on the dynamic alignment center coordinates. Based on the lateral force vector, an inverted conical spiral descent trajectory is generated. The gripper of the robotic arm is controlled to send the fiber optic pigtail connector into the guide port of the target fiber optic slot along the inverted conical spiral descent trajectory. After the fiber optic pigtail connector enters the guide port of the target fiber slot, a force-displacement curve is plotted. When the force-displacement curve shows a resistance peak point, the radius of curvature of the curve in the neighborhood of the resistance peak point is calculated. The radius of curvature of the curve is compared with a preset radius of curvature threshold to determine the authenticity of bottoming out. When a false bottoming out is determined, the side projection chord length is obtained, the under-insertion depth is inverted based on the side projection chord length, and high-frequency micro-oscillation compensation insertion is performed according to the under-insertion depth until the force-displacement curve shows the true locking characteristics.

2. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 1, characterized in that, The method for obtaining the polygonal outline of the irregular remaining holes includes: Semantic segmentation is performed on the top view image to distinguish the passable area of ​​the target fiber optic slot opening from the obstruction area of ​​the surrounding cables. The outer boundary of the passable area is extracted and fitted into an irregular residual hole polygonal contour. The coordinates of each vertex of the irregular residual hole polygonal contour are stored.

3. The end-effector pose dynamic compensation method based on the fusion of visual servoing and force feedback according to claim 2, characterized in that, The obtained dynamic alignment center coordinates include: Read the nominal length and nominal width values ​​of the cross-section of the fiber optic pigtail connector, calculate the aspect ratio of the cross-section of the fiber optic pigtail connector, solve for the maximum inscribed rectangle within the irregular remaining hole polygonal outline with the aspect ratio of the cross-section of the fiber optic pigtail connector as the constraint condition, and extract the center coordinates of the maximum inscribed rectangle as the dynamic alignment center coordinates.

4. The end-effector pose dynamic compensation method based on the fusion of visual servoing and force feedback according to claim 3, characterized in that, The process of obtaining the lateral force vector includes: When the length of the short side of the largest inscribed rectangle meets the safety margin requirement, the robotic arm is controlled to translate to directly above the dynamic alignment center coordinates; from directly above the dynamic alignment center coordinates, the Z-axis descent is executed. During the Z-axis descent, the lateral force components in the X-axis direction and the Y-axis direction are detected in real time by a six-dimensional force sensor, and the lateral force components in the X-axis direction and the Y-axis direction are combined into a lateral force vector.

5. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 4, characterized in that, The requirement to meet the safety margin means that the length of the short side of the maximum inscribed rectangle is greater than or equal to the sum of the nominal width of the cross-section of the fiber optic pigtail connector and the preset safety margin.

6. The end-effector pose dynamic compensation method based on the fusion of visual servoing and force feedback according to claim 5, characterized in that, The generation of the inverted conical spiral descent trajectory includes: The resultant force amplitude of the lateral force vector is compared with a preset compliance threshold. When the resultant force amplitude of the lateral force vector exceeds the preset compliance threshold, the gripper of the control robot arm generates a horizontal translation component along the normal direction of the lateral force vector. The displacement generated by the Z-axis actuator continuously descending in the vertical direction is recorded as the Z-axis descent component. The horizontal translation component and the Z-axis descent component are superimposed to form an inverted conical spiral descent trajectory.

7. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 6, characterized in that, The guide port for inserting the fiber optic pigtail connector into the target fiber optic slot includes: The fiber optic pigtail descends along an inverted conical spiral trajectory until it enters the guide port of the target fiber optic slot at the bottom of the fiber optic pigtail connector. When a rigid wall constraint resistance is detected in the Z-axis direction and the resultant force amplitude of the lateral force vector drops to near zero, it is determined that the fiber optic pigtail connector has entered the guide port of the target fiber optic slot. Near zero means that the fluctuation value drops to less than the preset lower limit threshold of the lateral force.

8. The end-effector pose dynamic compensation method based on the fusion of visual servoing and force feedback according to claim 7, characterized in that, The force-displacement curve plotting includes: After the fiber optic pigtail connector enters the guide port of the target fiber slot, the Z-axis actuator is controlled to continue to press down and advance at a constant speed in the vertical direction, and the cumulative downward displacement of the Z-axis and the insertion resistance in the Z-axis direction are recorded. The force-displacement curve is plotted point by point with the cumulative downward displacement of the Z-axis as the abscissa and the insertion resistance in the Z-axis direction as the ordinate.

9. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 8, characterized in that, The calculation of the curve curvature radius of the neighborhood of the resistance peak point includes: extracting curve data segments within a preset window width range before and after the resistance peak point, performing second derivative calculation on the curve data segments, and obtaining the curve curvature radius of the neighborhood of the resistance peak point.

10. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 9, characterized in that, The determination of whether the bottom has been reached includes: When the radius of curvature of the curve is less than the preset radius of curvature threshold, the resistance peak point is determined to correspond to a real latching event; when the radius of curvature of the curve is greater than or equal to the preset radius of curvature threshold, the resistance peak point is determined to correspond to a false bottoming event, the downward action of the Z-axis actuator is frozen, and the current cumulative downward displacement of the Z-axis is marked as the false bottoming frozen displacement.

11. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 10, characterized in that, The process of obtaining the side projection chord length includes: In the pseudo-bottoming and freezing state, the side image of the fiber optic pigtail connector is acquired from a preset oblique viewing angle. The trapezoidal projection contour of the fiber optic pigtail connector exposed outside the target fiber slot is identified from the side image. The side pixel length extending along the height direction of the fiber optic pigtail connector in the trapezoidal projection contour is extracted and recorded as the side projection chord length.

12. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 11, characterized in that, The under-interpolation depth inversion based on lateral projection chord length includes: An insertion depth mapping model is constructed, and the side projection chord length is input into the insertion depth mapping model to obtain the under-insertion depth of the fiber optic pigtail connector from the actual latching position.

13. The end-effector pose dynamic compensation method based on visual servoing and force feedback fusion according to claim 12, characterized in that, The high-frequency micro-oscillation compensation insertion includes: Based on the under-insertion depth control, the Z-axis actuator performs high-frequency micro-oscillation with a preset micro-oscillation amplitude and preset micro-oscillation frequency on the basis of the false bottoming-freezing displacement. When the under-insertion depth approaches zero and the radius of curvature of the curve in the neighborhood of the newly appeared resistance peak point is less than the preset radius of curvature threshold, it is determined that the fiber optic pigtail connector has reached the true locking position, the high-frequency micro-oscillation stops, the force is recovered and the clamp is released to complete the insertion operation. Approaching zero means that the monitored current under-insertion depth is less than the preset under-insertion depth convergence threshold.

Images