Self-adaptive shape following control method for multi-fingered dexterous hand facing soft and slippery pig liver grabbing

Abstract

The application discloses a kind of multi-fingered dexterous hand self-adaptive conformal control methods for soft slippery pig liver grabbing, to solve the problem of grabbing breakage and slip caused by extremely wet slippery and easily damaged pig liver, the application first constructs and visual guidance through orthogonal coordinate system, establish the preliminary grabbing configuration adapted to the irregular contour of pig liver;Then the dexterous hand is configured as low impedance state, drive finger along the normal compliance closure of pig liver surface, use contact reaction force to induce passive slip and roll of finger palm, realize the geometric conformation from point contact to surface contact;Again, through visual geometric shape closure criterion (fingertip depth across equator and envelope centroid) Trigger smooth variable stiffness locking based on balance point reset, the flexible posture is solidified as rigid grabbing structure in situ, and the dependence on friction is converted into geometric constraint.The application overcomes the risk of low friction slip under the premise of avoiding tissue stress damage, realizes the stable and non-destructive grabbing of soft slippery pig liver.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of robot dexterity manipulation, soft tissue biomechanics and intelligent control, and specifically to an adaptive conformal control method for a multi-finger dexterity hand for grasping soft, slippery pig liver. Background Technology

[0002] In the food industry and nutrition, pork liver, with its rich nutritional value and widespread market demand, has become an important part of the human diet. However, in the field of automated pork processing, the delicate handling of pork liver is an extremely challenging process. Pork liver possesses the dual characteristics of being extremely slippery and fragile: on the one hand, the tissue of pork liver has significant viscoelasticity and large deformation characteristics, and its shape will undergo drastic nonlinear deformation under external force; on the other hand, its surface is covered with mucus, resulting in an extremely low dynamic friction coefficient that changes nonlinearly with contact time and pressure, making it extremely easy to slip during the grasping process.

[0003] Existing dexterous hand grasping control methods have significant shortcomings in addressing these characteristics: When using position control or single-dimensional torque control, rigid fingertips cannot dynamically adapt to the nonlinear deformation of the liver tissue, which is highly deformable. This results in contact patterns often limited to discrete point or line contacts. This geometric mismatch prevents the grasping load from being effectively distributed over a large surface area, causing the local pressure at the contact point to momentarily exceed the yield strength of the liver surface, leading to damage to the liver tissue. Furthermore, relying solely on torque control cannot sensitively detect the precursors of microscopic slippage on the slippery surface of the liver. Once the target object exceeds the static friction limit and relative slippage occurs, the dexterous hand often fails to grasp due to the lack of an active geometric conformation and reconstruction mechanism.

[0004] Existing technologies also include methods for grasping pose planning based on 3D point clouds (such as CN121223809A), but their core still belongs to the model-driven framework of "planning-execution-correction," relying on prior modeling of the object's geometry and mechanical parameters. This cannot cope with the dual extreme characteristics of pig liver, such as its extremely slippery surface and highly compressible tissue. In the traditional approach, "pressure increase to prevent slippage" will lead to tissue damage, while "pressure reduction to preserve tissue" cannot overcome slippage. This contradiction has not yet been effectively resolved. Summary of the Invention

[0005] This invention addresses the problems in existing technologies for grasping soft, slippery pig liver, such as the "geometric mismatch" and "stress concentration" caused by rigid contact leading to tissue damage, and the slippage problem caused by the low frictional properties of the surface. This invention provides a multi-fingered dexterous hand adaptive conformal control method and system for grasping soft, slippery pig liver. The aim is to establish a spatial envelope relationship with the irregular contour of the pig liver through "visually guided geometric pre-configuration" based on orthogonal coordinate system correction, combined with a "passive conformal" strategy under low impedance contact. This allows the multi-fingered dexterous hand to actively adapt to the viscoelastic nonlinear deformation of the pig liver tissue, maximizing the contact fit between the fingertips and the pig liver surface. Furthermore, a "visual geometric closure criterion" is used to monitor the envelope state of the pig liver in real time and trigger variable stiffness locking based on impedance balance point reset. This overcomes the risk of low-friction slippage caused by surface mucus while avoiding stress damage to the pig liver epidermis and sudden stiffness shocks, achieving stable grasping of soft, slippery pig liver.

[0006] An adaptive conformal control method for a multi-finger dexterous hand for grasping soft, slippery pig liver includes the following steps: Step 1: Perform decentralized principal component analysis on the pig liver point cloud, and construct an absolutely orthogonal grasping reference coordinate system by combining the gravity anti-direction vector through continuous double vector cross product; at the same time, calculate the maximum curvature of the surface based on the local quadratic surface fitting of the point cloud, dynamically map the initial bending angle of each finger, and drive the multi-finger dexterous hand to form a preparatory grasping configuration adapted to the irregular contour without touching the pig liver. Step 2: Configure the joint controllers of the multi-finger dexterous hand to a low-impedance control state, drive the fingers to perform low-speed compliant closure along the local normal of the pig liver surface; use the contact reaction force of the pig liver surface on the fingertip to induce the finger to produce a passive sliding-rolling composite adaptive displacement until the dual criteria of force convergence and geometric alignment are met, that is, the contact force of the fingertip normal converges stably and the fingertip surface is aligned with the local normal vector of the pig liver surface, realizing the adaptive geometric conformal from discrete point contact to continuous surface contact; Step 3: During the conformal contact process, the spatial enclosure relationship between the fingertips of the multi-fingered dexterous hand and the centroid of the pig liver is monitored in real time using a visual geometric monitoring algorithm. When both the depth crossing constraint and the centroid coverage constraint are satisfied, it is determined that the multi-fingered dexterous hand has formed a stable structural constraint cage, triggering the shape closure ready signal. Among them, the depth crossing constraint is that the vertical height of the fingertips of at least three effective contacting fingers is lower than the preset depth of the equatorial surface of the maximum contour of the pig liver, and the centroid coverage constraint is that the smallest convex polygon formed by the projection points of the fingertips that satisfy the above preset depth in the horizontal plane contains the projection point of the centroid of the pig liver. Step 4: Once the shape-closure ready signal is received, the controller executes the in-situ variable stiffness locking strategy. First, it combines the current low stiffness matrix with the actual contact force to reset the impedance model equilibrium point to calculate the new desired equilibrium position, ensuring a seamless and smooth transition of the contact force. Then, it uses a smooth transition curve to gradually increase the stiffness coefficient in the impedance control parameters from a low value to a high stiffness holding value, solidifying the flexible conformal posture in-situ into a rigid envelope structure. It uses the geometric interference formed by the finger skeleton and the surface of the pig liver to resist inertial and gravitational disturbances, achieving stable gripping of the soft and slippery pig liver.

[0007] Furthermore, step 1 further includes: Point cloud data of pig liver surface was collected using an RGB-D depth vision sensing device. A time-domain weighted fusion algorithm based on HSV color space saturation confidence was used to suppress high reflective noise and generate high-fidelity 3D point cloud to construct a static benchmark model. Based on the point cloud statistical features, the geometric centroid of the point cloud is extracted and decentralized principal component analysis is performed to extract the direction vector of the maximum eigenvalue. Combined with the system gravity inverse direction vector, an absolutely orthogonal grasping reference coordinate system that satisfies the right-hand rule is constructed through continuous quadratic vector cross product. The three-dimensional point cloud is projected onto the horizontal plane, and the maximum contour width of the pig liver is calculated using the convex hull calculation algorithm. The size of the safe inner envelope space is set by combining the preset safety gap coefficient and soft tissue compression compensation amount. The maximum curvature of the pig liver surface is calculated using a spatial neighborhood search algorithm and a local quadratic surface fitting method. The initial pre-bending angle of the finger is then dynamically mapped based on the maximum curvature.

[0008] Furthermore, the low impedance control state in step 2 specifically refers to: The stiffness matrix in the impedance control parameters is set to a diagonal matrix, where the translational stiffness coefficient of the multi-fingered dexterous hand along the fingertip feed direction and the surface tangent is set to... Set the angular stiffness coefficient of the rotation dimension to This allows for yielding displacement and passive roll of the fingertips; and the damping matrix is ​​set to be in an overdamped state with a damping ratio of .

[0009] Furthermore, the condition for determining in step 2 that the fingertip surface and the local normal vector of the pig liver surface are aligned is: Force balance criterion: The normal contact force of the fingertip converges stably to And the modulus of the rate of change of contact force is less than Geometric alignment criterion: The dot product of the fingertip unit normal vector calculated by forward kinematics and the liver unit normal vector of the current contact point as shown by visual feedback. .

[0010] Further, the slip-roll composite adaptive displacement is as follows: the contact reaction force applied by the pig liver surface to the fingertip is naturally decomposed into a normal component perpendicular to the fingertip surface and a tangential component along the contact tangential plane; under the action of the low stiffness impedance control law, the non-zero tangential component generates a lateral correction acceleration in the fingertip coordinate system, driving the fingertip to produce tangential slip along the direction of the decrease in curvature of the pig liver surface; at the same time, under the action of the contact torque generated by the contact point deviating from the geometric center of the fingertip, the fingertip is induced to passively roll around the contact point until the fingertip surface is aligned with the local normal vector of the pig liver surface.

[0011] Furthermore, the preset depth in step 3 is 5mm.

[0012] Furthermore, the variable stiffness locking strategy in step 4 further includes: Before performing stiffness enhancement, a balance point reset operation is performed: the actual angles of each joint and the actual position of the fingertip in Cartesian space at the current moment are read. and the current stable contact force Calculate the new desired equilibrium position including force maintenance compensation. ,in It is the inverse matrix of the low stiffness matrix, so that the normal compressive force of the fingertip on the pig liver is maintained within the compliant safety value range set in step 2 at the moment of stiffness switching. Within the set transition time window, a smooth transition curve is used to gradually increase the diagonal elements of the stiffness matrix from low stiffness values ​​to high stiffness retention values. .

[0013] Furthermore, the smooth transition curve is an S-shaped curve or a cubic polynomial curve.

[0014] Furthermore, step 3 also includes a replanning mechanism: If the visual feedback judgment fails to meet the geometric closure trigger condition within the preset maximum detection period, the replanning interrupt service is triggered: the controller stops fingertip feeding, switches back to high impedance position control mode, drives the multi-finger dexterous hand to lift to the initial hovering height, and triggers the vision system to re-execute step 1 to obtain the latest pose and shape of the pig liver, and replans the grasping configuration.

[0015] Furthermore, the multi-finger dexterous hand is a rigid multi-finger dexterous hand with five fingers, and each joint is equipped with a position sensor and a force sensor.

[0016] The beneficial effects of this invention are as follows: It abandons traditional rigid position control and adopts a low-stiffness impedance contact strategy to endow the rigid multi-finger dexterous hand with physical compliance. Through a dual convergence criterion of force and geometry, it guides the fingertips to undergo passive "sliding-rolling" displacement, enabling them to adapt to the viscoelastic nonlinear deformation of pig liver tissue, thereby maximizing the geometric fit between the fingertips and the irregular surface of the pig liver. This mechanism transforms the contact form from "discrete point contact," which easily causes stress concentration, into "continuous surface contact." Under the same grasping load, it significantly reduces the contact pressure per unit area, effectively avoiding puncture or compression damage to the pig liver tissue caused by contact singularities. Visual geometric feedback (i.e., dual constraints of depth crossing and core coverage) is used as the threshold judgment condition for control mode transition, and the envelope state is detected in real time, so that the grasping stability is no longer limited by the low friction of the pig liver surface. Under extremely slippery conditions, as long as the geometric envelope structure cage remains intact, physical barriers can be used to ensure that the object does not shift or slip, solving the problem of "pressure-induced anti-slip and pressure-induced damage" in traditional control, and realizing stable grasping of soft and slippery pig liver; By combining the low-stiffness inverse matrix with the balance point reset of the actual contact force, and smoothing the variable stiffness curve (S-shaped or cubic polynomial), the impact force caused by sudden stiffness changes is avoided from crushing soft tissue; at the same time, a replanning mechanism is set up to automatically lift and replan when the shape closure fails, thereby improving the gripping robustness. Attached Figure Description

[0017] Figure 1 This is a complete flowchart of the present invention; Figure 2 This is a diagram showing the passive conformal and force analysis of the multi-fingered dexterous hand in low-resistance mode on the soft and smooth surface of pig liver according to the present invention. Figure 3 This is a schematic diagram illustrating the geometric constraint principle of the present invention for determining the shape closure of soft, smooth pig liver under visual geometric feedback. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings. Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The directional terms such as left, center, right, top, and bottom in the embodiments of the present invention are only relative concepts or referenced to the normal use state of the product, and should not be considered restrictive.

[0019] System components: This embodiment employs a two-layer control architecture of "industrial computer-robot controller". The vision recognition system running on the industrial computer drives an RGB-D depth camera (preferably Intel RealSense D455) fixed to the end effector of the dexterous hand via a USB 3.0 data interface. The robot controller (such as a real-time controller based on the EtherCAT bus) is responsible for driving the movement of each joint of the dexterous hand and the robotic arm. Each joint of the dexterous hand is equipped with a force sensor or current feedback loop for impedance control. The multi-fingered dexterous hand is a rigid five-fingered dexterous hand, with each joint equipped with a position sensor and a force sensor.

[0020] An adaptive conformal control method for multi-finger dexterous hand grasping soft, smooth pig liver, such as Figure 1 As shown, it includes the following steps: Step 1: Preliminary grasping configuration matching based on visual features and orthogonal coordinate system The visual recognition system collects the depth information of the soft, smooth pig liver and executes an anti-reflective fusion algorithm to calculate the geometric feature parameters of the target point cloud and send them to the robot controller. The robot controller calls the inverse kinematics model based on the received feature parameters, drives the multi-fingered dexterous hand to adjust the palm posture and finger angle, and forms a pre-grasping configuration that matches the irregular outer contour of the pig liver without contacting the target.

[0021] Specifically, step 1 further includes: High-fidelity point cloud acquisition: Addressing the lack of structured light depth due to the "high reflectivity and strong mucus" characteristics of pig liver surface, a vision recognition system running on an industrial control computer drives an RGB-D depth camera fixed to the end effector of a dexterous hand via a USB 3.0 data interface to continuously acquire N frames of depth data streams and RGB images in a temporal multi-frame sliding window mode. The system reads the video memory data in real time, extracts the HSV saturation components of each pixel in the RGB image, and substitutes the depth stream data from multiple consecutive frames into a "confidence-based temporal weighted average filtering model" for fusion calculation. The total number of frames within the sampling window is set to... (Preferred, this embodiment sets) For coordinates in the image, For any pixel, its final depth value after fusion processing Calculate using the following formula: in, For the first The raw depth input value captured in the frame; For the first The coordinates in the frame are The weight coefficients corresponding to the pixels. To effectively suppress specular noise, its calculation formula is set as follows: ;in This represents the normalized saturation of the pixel in the HSV color space. The saturation gain coefficient is set (preferably, in this embodiment, it is set to...). After completing the calculations using the above algorithm, a smooth 3D point cloud with suppressed specular noise is generated. Then, the radius filter in the PCL point cloud library is called to remove outliers, and the moving least squares method is used to repair data holes. The processed static point cloud data and its absolute position are locked as a static benchmark model and stored in memory as the sole benchmark for geometric analysis. Establishing a grasping reference coordinate system: To plan the grasping posture, the system needs to establish a grasping reference coordinate system based on the pig liver itself, traversing the filtered valid point cloud data, and setting the total number of points in the valid point cloud data to be... Extract the first The spatial coordinates of the sampling points are Based on the first-order moment principle of point clouds, the spatial coordinates of all points are substituted to solve for the geometric centroid coordinates of the pig liver point cloud. : Obtain the center of mass Next, all valid point cloud coordinates are decentered by subtracting the centroid coordinates. Then, principal component analysis (PCA) is performed to construct the covariance matrix of the centered point cloud and decompose the eigenvectors. The direction vector corresponding to the largest eigenvalue is extracted as the initial ordinate vector. To ensure the orthogonality of the axes of the grasping coordinate system, the gravity in the opposite direction vector set by the system is used as the Z-axis reference vector. ;calculate and The result of the cross product of the vectors is used as the Y-axis reference vector. (Right now ); and then calculate and The result of the vector cross product is used as the corrected X-axis reference vector. (Right now Finally, regarding , , Unitize, with the center of mass Using the origin as the coordinate origin, construct a unique grasping reference coordinate system that satisfies the right-hand rule and is absolutely orthogonal.

[0022] Calculate the maximum contour width and finger pre-bending angle: In this coordinate system, further calculate the key geometric commands that determine the dexterous hand configuration; to determine the finger opening range, the system projects the 3D point cloud onto the XOY plane of the grasping reference coordinate system, and uses a convex hull calculation algorithm (in this embodiment, the Graham scan algorithm is preferred) to construct the convex hull of the 2D projected point set, setting the total number of vertices of the convex hull. Extract any two non-repeating vertices on the convex hull contour. and (satisfy Coordinate input values and Substituting this into the Euclidean distance formula and performing traversal calculations, we search for the upper limit of the physical size of the pig liver on the grasping cross section. : Secondly, to calculate the maximum surface curvature to assist in setting the initial bending posture of the finger, the industrial control computer traverses each point in the point cloud and uses a spatial neighborhood search algorithm (preferably the KD-Tree algorithm in this embodiment) to find its nearest neighbor point set. Based on this neighbor point set, the normal vector of the current point is solved through principal component analysis, and with this point as the origin, the normal vector is... A local tangent space coordinate system is constructed using axes. After transforming neighborhood points to this local coordinate system, they are substituted into the local quadratic surface model for least-squares fitting. in, This represents the height fitted value in the local tangent space coordinate system; To obtain the surface fitting coefficients; using the coefficients The two principal curvatures at that point can be directly calculated from the second-order Hessian matrix. and The calculation formula is as follows: The larger of the two absolute values ​​is taken as the local curvature value of the sampling point. The global maximum value obtained after traversing all values ​​is the maximum curvature of the surface. ; Based on the maximum curvature of this surface Set threshold parameters and (Preferredly, in this embodiment, the following is taken) , Dynamically map the initial bending angle of the finger. The mapping strategy is configured as follows: when At that time, it indicates that the surface outline of the pig liver is relatively sharp, setting (Preferred) ); when At this time, it indicates that the curvature of the pig liver surface contour is moderate, and the setting is appropriate. (Preferred) ); when At this time, it indicates that the surface contour of the pig liver is relatively flat, and the setting is... (Preferred) This allows the initial curvature of the fingertip to adapt to the local geometry of the pig liver surface, increasing the contact area and reducing the pressure per unit area. Generate the pre-grabbing configuration: The industrial control computer sets the safety clearance coefficient. (The preferred value range is 0.05 to 0.15, and in this embodiment, it is 0.08), and a soft tissue compression compensation amount is introduced. Calculate the target width of the inner envelope space. : ; The calculated "target envelope width" "Finger pre-bending angle" The commands “grab reference coordinate system pose” and “grab reference coordinate system pose” are packaged into control instructions and sent to the robot controller via TCP / IP industrial Ethernet; After receiving the above instructions, the robot controller substitutes them into the inverse kinematics model of the dexterous hand based on DH parameters, calculates the target motor angles corresponding to each joint, and controls the dexterous hand to concurrently perform multi-dimensional posture reconstruction via the EtherCAT real-time bus: on the one hand, it drives the end effector to adjust the palm normal vector to be parallel to the Z-axis of the grasping coordinate system; on the other hand, it adjusts the finger spacing angle to make the inner envelope space size equal to And drive the proximal joints of each finger to bend to the set position. To adapt to the surface curvature; The robot controller moves the robotic arm, carrying a dexterous hand, directly above the pig liver, keeping the fingertips at the highest point of the liver's surface. Hovering distance (preferably, Set to 10-15mm), complete the geometric pre-configuration matching; Step 2: Passive Follower with "Slip-Roll" in Low Impedance Mode Based on the pre-grasping configuration generated in step 1, to address the stress concentration and tissue damage issues that easily arise when rigid mechanical fingertips directly contact pig liver, which has high viscoelasticity and a slippery surface, the joint controllers of the multi-fingered dexterous hand are configured to a low-impedance control state. More specifically, this step establishes a second-order impedance dynamics model of the multi-fingered dexterous hand's end effector in Cartesian coordinates. This model describes the dynamic equilibrium relationship between fingertip position deviation and contact force. in, , , These are the desired inertia matrix, damping matrix, and stiffness matrix, respectively. These are the desired position, velocity, and acceleration vectors of the preset trajectory, respectively. These are the actual position, velocity, and acceleration vector of the fingertip, respectively. The contact reaction force applied to the surface of the pig liver by the fingertip; To achieve a "flexible conformal" shape for the soft, smooth pig liver, the impedance parameters need to be configured specifically: the stiffness matrix... Set as a diagonal matrix and decouple its main diagonal elements to include the linear stiffness coefficients along the fingertip feed direction (i.e., the normal) and the surface tangent. Set as Within an extremely low range, the angular stiffness coefficient of the rotation dimension is simultaneously set to an extremely low threshold (preferably). This design aims to allow the fingertip to produce significant yield displacement and roll even under weak reaction forces and contact torques from pig liver. Simultaneously, a damping matrix is ​​designed to address the characteristics of the mucus layer on the surface of pig liver, which easily leads to contact instability and high-frequency slippage. In an overdamped state, select the damping ratio. The damping parameter is calculated using the following formula to absorb the impact energy at the moment of contact: After determining the aforementioned expectation matrix, the control system acquires the joint angles of the current multi-fingered dexterous hand in real time. And calculate the corresponding Jacobian transpose matrix. The system uses the Jacobian transpose matrix to map the solved Cartesian space virtual force into the output torque of each joint motor. This drives the movement of the physical joint, and its impedance control law expression is: In the conformal control stage, based on the point cloud features of the pig liver surface extracted in step 1, the local normal vectors of the contact areas corresponding to each fingertip are calculated. Planning along the fingertips direction The low-speed approach to pig liver; using a six-dimensional force sensor integrated into the fingertip to monitor the contact state in real time at a frequency of 1kHz, when the contact force modulus is detected... Exceeding the trigger threshold At that time, the system determines that contact has occurred.

[0023] At this point, the viscoelastic reaction force of the pig liver tissue This disrupts the equilibrium of the impedance equation, especially at low stiffness. Under its influence, the actual trajectory of the fingertip is forced. Deviating from rigid planning trajectory Specifically, the contact reaction force exerted on the surface of the pig liver by the fingertip. It will be naturally decomposed into a normal component force perpendicular to the surface of the fingertip. and the tangential component along the contact tangential plane Under the action of the low-stiffness impedance control law, the non-zero tangential component force This generates a lateral corrective acceleration within the fingertip coordinate system, driving the fingertip to slide tangentially along the direction of decreasing curvature of the pig liver surface. Simultaneously, the contact torque generated due to the contact point deviating from the fingertip's geometric center acts on the rotational dimension of the impedance dynamics model. Under the aforementioned extremely low angular stiffness coefficient, this contact torque induces the fingertip to passively roll around the contact point until the fingertip surface aligns with the local normal vector of the pig liver surface. This passive conformal process continues until the following dual convergence criteria are met: The first is the force balance criterion, which requires the normal contact force of the fingertip. Stable convergence to the compliant maintenance range And its normal contact force change rate This indicates that the contact has stabilized and there has been no destructive compression. Second, the geometric alignment criterion, calculated through forward kinematics, is the unit normal vector at the fingertip. The normal vector of the pig liver unit at the current contact point of visual feedback The directional consistency constraint must be satisfied, and its mathematical expression is: That is, the cosine of the angle between the two is greater than The corresponding angular deviation is less than This "slip-roll" composite motion, driven by force perception deviation, will continue, automatically seeking the posture with the highest fit to the pig liver surface. When the above criteria are met, it is determined that the fingertip surface has completely conformed to the irregular deformation of the pig liver surface, achieving adaptive fit from "point contact" to "surface contact." The controller then records the current joint angles and end-effector poses as the reference configuration for subsequent variable stiffness grasping. Figure 2 As shown in the figure, the sliding and rolling motion of the fingertip under the action of normal and tangential components of force is illustrated.

[0024] Step 3: During the conformal contact process, monitor in real time the spatial enclosure relationship between the fingertips of the multi-fingered dexterous hand and the centroid of the pig liver; specifically: While the multi-fingered dexterous hand performs low-impedance passive conformal control, a visual geometric monitoring algorithm is run in parallel to establish reliable mechanical constraints on the extremely low friction coefficient surface of the pig liver. This algorithm aims to determine whether the current grasping configuration satisfies the physical condition of "shape closure." This monitoring process directly uses the local grasping reference coordinate system established in step 1 (i.e., the pig liver centroid calculated in step 1). (with the origin as the coordinate system and the Z-axis perpendicular to it), using an RGB-D depth camera to... The system captures scene point clouds in real time at high frequency and combines this with real-time calculation of the spatial coordinates of each fingertip center using the kinematics of the dexterous hand. Perform geometric calculations. To quantify the effectiveness of the fingertip's envelope on the irregular contour of the pig liver, the algorithm first... The real-time pig liver point cloud is sliced ​​and scanned along the axial direction. The point cloud slice with the largest contour cross-sectional area is extracted and its plane is defined as the "maximum contour equatorial plane". And read the height value of the plane in the Z-axis. Based on this, the system sets a dual geometric constraint criterion for determining the formation of a "stable structural constraint cage". A shape closure is considered achieved only when both of the following conditions are met simultaneously: The first criterion is the depth-crossing constraint, which aims to use the physical interference formed by the finger skeleton to suppress the slippage and multi-axial flipping tendency of the pig liver along the direction of gravity (i.e., the negative Z-axis). The system detects the Z-axis coordinates of each fingertip. Based on the spatial geometry principle of "three points determine a plane," it requires that the vertical height of at least three effectively contacting fingertips is lower than the equatorial plane of the pig liver, that is, to ensure that the fingertips have substantially entered the contraction area of ​​the lower half of the pig liver, thereby constructing a stable "triangular base" structure in space; its mathematical expression is: in, A set of finger indices that satisfy the depth-crossing condition; The number of elements in the set; For the first The fingertip is grasping the real-time spatial coordinates of the reference coordinate system along the Z-axis. This is a safety redundancy depth parameter, i.e., a preset depth, set to compensate for point cloud measurement deviations and improve the reliability of mechanical structure positioning. (Preferably, in this embodiment, it is taken as...) ).

[0025] The second criterion is the heart region coverage constraint, which aims to constrain the escape and rotation of the pig liver in the XY plane. The system projects the coordinates of all fingertips that satisfy the above depth constraints onto the XY plane, constructing the minimum convex polygon of the fingertip projection point set. ,Right now At the same time, the core of the pig liver Projected onto the same plane (due to the definition of the coordinate system, the centroid projection point) Coordinates are always The system uses a ray casting algorithm to determine whether the centroid projection point is located inside the convex polygon formed by the fingertip projection, i.e., whether it satisfies the geometric inclusion relationship. When the above two geometric criteria are in continuous Within a frame image period (preferably, When all three conditions are met with a true value, it indicates that the multi-fingered dexterous hand has successfully constructed a closed geometric cage around the pig liver in three-dimensional space, such as... Figure 3As shown in the figure, the relationship between the fingertip equatorial plane crossing and the centroid projection is illustrated. At this time, the six-dimensional degrees of freedom of the pig liver have been effectively constrained by the physical barrier formed by the finger skeleton. It is determined that a stable structural constraint cage has been formed, and the system then outputs a "shape closed and ready" signal to trigger the subsequent variable stiffness locking process.

[0026] It also has a replanning mechanism: If the maximum detection cycle specified in step 3 (Preferably set as) Within this range, the visual feedback system consistently determines that the "shape closure" condition (i.e., continuous) is not met. If a frame meets the criteria (e.g., a pig liver accidentally slides extensively due to excessive slipperiness), the system will trigger a replanning interruption service. The robot controller immediately stops the fingertip feed, switches back to the high-impedance position control mode, drives the dexterous hand to lift vertically along the Z-axis to the initial hovering height, and triggers the vision system to re-execute step 1 to obtain the latest pose and shape of the pig liver, re-plan the grasping configuration, and avoid blind closing that could lead to grasping failure or equipment damage.

[0027] Step 4: Dynamic variable stiffness locking and stable gripping based on equilibrium point reset In practical implementation, once the control system receives the "shape-closed ready" trigger signal output in step 3, it indicates that the dexterous hand has constructed a safe geometric envelope in low-impedance mode. The system immediately terminates the passive conformal process in step 2 and enters the dynamic variable stiffness locking stage. The core strategy of this stage is to perform "in-situ solidification" control, that is, to smoothly transition the physical characteristics of the dexterous hand's end from "flexible compliance" to "rigid retention" while maintaining the current contact pose between the fingertip and the surface of the pig liver without abrupt changes.

[0028] To prevent a sudden surge in contact force due to an increase in stiffness parameters, which could crush soft tissue, the controller first performs an impedance model equilibrium point reset operation: real-time reading of the actual angles of each joint and the actual position of the fingertip in Cartesian space. and the current stable contact force vector To prevent contact force loss or abrupt changes during stiffness transitions, the controller calculates a new desired equilibrium position that includes 'force maintenance compensation'. The calculation formula is as follows: in, This is the low-stiffness matrix at the end of step 2, following the conformal control. The purpose of this operation is to reset the equilibrium point of the impedance model so that at the initial instant of stiffness parameter switching, the theoretically calculated force output by the model is... Strictly equal to the current actual contact force This achieves a smooth transition of contact force, allowing the normal compressive force of the fingertips on the pig liver to continue to be maintained within the compliant safety range set in step 2. Inside, to avoid generating additional clamping work.

[0029] After the balance point is reset, the system dynamically modulates the impedance parameters based on time-flow. Within the set transition time window... Inside (preferably, set) The stiffness matrix is ​​smoothed using a cubic polynomial transition curve. By gradually increasing the diagonal elements from low stiffness values ​​to high stiffness-maintaining values, the impact of sudden stiffness changes can be reduced. Let this high stiffness matrix be... The high stiffness coefficient scalar on its main diagonal is set as (The preferred value range is) The expression for the evolution of its stiffness over time is: With a smooth, nonlinear increase in stiffness, the joint servo stiffness of each finger of the dexterous hand is significantly enhanced while maintaining the existing conformal surface, thereby constructing a rigid spatial envelope structure in physical space that highly matches the morphological characteristics of the pig liver surface. This structure transforms the originally loose, flexible contact into a rigid constraint with high resistance to disturbance. In this high-stiffness mode, the grasping stability no longer depends on the mucous friction force on the pig liver surface, but is transformed into the geometric constraint force provided by the "depth crossing" and "heart domain coverage" described in step 3. At this point, even if the pig liver experiences a slight downward displacement due to gravity, the fingertips are already located on the equatorial plane. Below and the stiffness has been increased to the locked value. The finger skeleton will generate a normal supporting force sufficient to balance gravity, according to Hooke's Law. : in, This represents the actual spatial position vector generated by the fingertip when the pig liver undergoes a slight slippage tendency.

[0030] Through the time-varying modulation mechanism of the aforementioned impedance parameters, the system successfully transforms the geometric fit advantage accumulated through "flexible conformal" into the mechanical advantage of "rigid locking." Ultimately, the control system instructs the robotic arm to maintain its current high-rigidity cage-like grasping configuration with limited lifting acceleration. (Preferred, limited) It performs vertical lifting and transporting actions; during this dynamic process, the rigid envelope cage constructed by the finger skeleton effectively resists the disturbance caused by inertial force, and completes the stable grasping of soft and slippery pig liver without applying destructive clamping force to the surface of the pig liver.

[0031] A multi-finger dexterous hand adaptive conformal grasping system includes: An RGB-D depth vision sensing module is used to acquire three-dimensional point cloud data of the surface of pig liver. Industrial PCs and robot controllers are used to perform visual feature extraction, geometric pre-configuration matching, low-impedance control parameter configuration, geometric closure determination, and variable stiffness locking control. The multi-finger dexterous hand body is equipped with an impedance controller at each joint, which can achieve passive conformal displacement in a low impedance state and maintain a rigid grasping configuration in a variable stiffness locked state. The force sensing module, integrated into the fingertip, is used to monitor contact force in real time and feed it back to the controller.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-finger dexterous hand adaptive conformal control method for grasping soft, slippery pig liver, characterized in that, Includes the following steps: Step 1: Perform decentralized principal component analysis on the pig liver point cloud, and construct an absolutely orthogonal grasping reference coordinate system by combining the gravity anti-direction vector through continuous double vector cross product; at the same time, calculate the maximum curvature of the surface based on the local quadratic surface fitting of the point cloud, dynamically map the initial bending angle of each finger, and drive the multi-finger dexterous hand to form a preparatory grasping configuration adapted to the irregular contour without touching the pig liver. Step 2: Configure the joint controllers of the multi-finger dexterous hand to a low-impedance control state, drive the fingers to perform low-speed compliant closure along the local normal of the pig liver surface; use the contact reaction force of the pig liver surface on the fingertip to induce the finger to produce a passive sliding-rolling composite adaptive displacement until the dual criteria of force convergence and geometric alignment are met, that is, the contact force of the fingertip normal converges stably and the fingertip surface is aligned with the local normal vector of the pig liver surface, realizing the adaptive geometric conformal from discrete point contact to continuous surface contact; Step 3: During the conformal contact process, the spatial enclosure relationship between the fingertips of the multi-fingered dexterous hand and the centroid of the pig liver is monitored in real time using a visual geometric monitoring algorithm. When both the depth crossing constraint and the centroid coverage constraint are satisfied, it is determined that the multi-fingered dexterous hand has formed a stable structural constraint cage, triggering the shape closure ready signal. Among them, the depth crossing constraint is that the vertical height of the fingertips of at least three effective contacting fingers is lower than the preset depth of the equatorial surface of the maximum contour of the pig liver, and the centroid coverage constraint is that the smallest convex polygon formed by the projection points of the fingertips that satisfy the above preset depth in the horizontal plane contains the projection point of the centroid of the pig liver. Step 4: Once the shape-closed ready signal is received, the controller executes the in-situ variable stiffness locking strategy; first, it combines the current low stiffness matrix with the actual contact force and performs impedance model equilibrium point reset to calculate the new desired equilibrium position, ensuring a seamless and smooth transition of the contact force; Subsequently, a smooth transition curve was used to gradually increase the stiffness coefficient in the impedance control parameters from a low value to a high stiffness retention value, thus solidifying the flexible conformal posture in situ into a rigid envelope structure. The geometric interference formed by the finger skeleton and the surface of the pig liver was used to resist inertial and gravitational disturbances, thereby achieving stable grasping of the soft and slippery pig liver.

2. The multi-finger dexterous hand adaptive conformal control method for grasping soft, smooth pig liver according to claim 1, characterized in that, Step 1 further includes: Point cloud data of pig liver surface was collected using an RGB-D depth vision sensing device. A time-domain weighted fusion algorithm based on HSV color space saturation confidence was used to suppress high reflective noise and generate high-fidelity 3D point cloud to construct a static benchmark model. Based on the point cloud statistical features, the geometric centroid of the point cloud is extracted and decentralized principal component analysis is performed to extract the direction vector of the maximum eigenvalue. Combined with the system gravity inverse direction vector, an absolutely orthogonal grasping reference coordinate system that satisfies the right-hand rule is constructed through continuous quadratic vector cross product. The three-dimensional point cloud is projected onto the horizontal plane, and the maximum contour width of the pig liver is calculated using the convex hull calculation algorithm. The size of the safe inner envelope space is set by combining the preset safety gap coefficient and soft tissue compression compensation amount. The maximum curvature of the pig liver surface is calculated using a spatial neighborhood search algorithm and a local quadratic surface fitting method. The initial pre-bending angle of the finger is then dynamically mapped based on the maximum curvature.

3. The multi-finger dexterous hand adaptive conformal control method for grasping soft, smooth pig liver according to claim 1, characterized in that, The low-impedance control state in step 2 is specifically as follows: The stiffness matrix in the impedance control parameters is set to a diagonal matrix, where the translational stiffness coefficient of the multi-fingered dexterous hand along the fingertip feed direction and the surface tangent is set to... Set the angular stiffness coefficient of the rotation dimension to This allows for yielding displacement and passive rolling of the fingertips; And the damping matrix is ​​set to be in an overdamped state, with a damping ratio of .

4. The multi-finger dexterous hand adaptive conformal control method for grasping soft, smooth pig liver according to claim 1, characterized in that, The condition for determining whether the fingertip surface and the local normal vector of the pig liver surface are aligned in step 2 is as follows: Force balance criterion: The normal contact force of the fingertip converges stably to And the modulus of the rate of change of contact force is less than Geometric alignment criterion: The dot product of the fingertip unit normal vector calculated by forward kinematics and the liver unit normal vector of the current contact point as shown by visual feedback. .

5. The multi-finger dexterous hand adaptive conformal control method for grasping soft, slippery pig liver according to claim 1, characterized in that, The slip-roll composite adaptive displacement is as follows: the contact reaction force applied to the fingertip by the pig liver surface is naturally decomposed into a normal component perpendicular to the fingertip surface and a tangential component along the contact tangential plane; under the action of the low stiffness impedance control law, the non-zero tangential component generates a lateral correction acceleration in the fingertip coordinate system, driving the fingertip to produce tangential slip along the direction of the decrease in curvature of the pig liver surface; at the same time, under the action of the contact torque generated by the contact point deviating from the geometric center of the fingertip, the fingertip is induced to passively roll around the contact point until the fingertip surface is aligned with the local normal vector of the pig liver surface.

6. The multi-finger dexterous hand adaptive conformal control method for grasping soft, smooth pig liver according to claim 1, characterized in that, The preset depth in step 3 is 5mm.

7. The multi-finger dexterous hand adaptive conformal control method for grasping soft, slippery pig liver according to claim 1, characterized in that, The variable stiffness locking strategy in step 4 further includes: Before performing stiffness enhancement, a balance point reset operation is performed: the actual angles of each joint and the actual position of the fingertip in Cartesian space at the current moment are read. and the current stable contact force Calculate the new desired equilibrium position including force maintenance compensation. ,in It is the inverse matrix of the low stiffness matrix, so that the normal compressive force of the fingertip on the pig liver is maintained within the compliant safety value range set in step 2 at the moment of stiffness switching. Within the set transition time window, a smooth transition curve is used to gradually increase the diagonal elements of the stiffness matrix from low stiffness values ​​to high stiffness retention values. .

8. The multi-finger dexterous hand adaptive conformal control method for grasping soft, smooth pig liver according to claim 7, characterized in that, The smooth transition curve is an S-shaped curve or a cubic polynomial curve.

9. The multi-finger dexterous hand adaptive conformal control method for grasping soft, slippery pig liver according to claim 1, characterized in that, Step 3 also includes a replanning mechanism: If the visual feedback judgment fails to meet the geometric closure trigger condition within the preset maximum detection period, the replanning interrupt service is triggered: the controller stops fingertip feeding, switches back to high impedance position control mode, drives the multi-finger dexterous hand to lift to the initial hovering height, and triggers the vision system to re-execute step 1 to obtain the latest pose and shape of the pig liver, and replans the grasping configuration.

10. The multi-finger dexterous hand adaptive conformal control method for grasping soft, slippery pig liver according to claim 1, characterized in that, The multi-finger dexterous hand is a rigid multi-finger dexterous hand with five fingers, and each joint is equipped with a position sensor and a force sensor.

Images