Round-like object grabbing method and system based on dexterous hand

Abstract

The invention provides a dexterous hand-based quasi-round object grabbing method and system, and relates to the technical field of fruit picking, the dexterous hand comprises at least three fingers with finger root rotating structures, and the finger tip of each finger is symmetrically provided with two pressure detection pieces; the round-like object grabbing method based on the dexterous hand comprises the steps that the ends of dexterous fingers make contact with the surface of a round-like object, pressure detection piece data and initial pose information on all the fingers are collected, and the pressure detection piece data form a pressure data set; according to the pressure data set, a description value corresponding to each finger tip is obtained, the estimated mass center of the quasi-circle object is obtained, and the description values are used for reflecting the deviation between the force application direction of the fingers and the normal direction of the surface of the quasi-circle object. According to the method, the finger tip pressure data and the initial pose information are collected, the description value reflecting the force application direction deviation and the estimated mass center are calculated according to the data, real-time accurate sensing of the contact force direction is achieved, and the problem that the force vector deviation cannot be sensed in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of fruit picking technology, and in particular to a method and system for grasping round objects based on dexterous hands. Background Technology

[0002] In recent years, driven by scale and consumption upgrades, the fruit industry has become one of the core sectors of the agricultural economy. According to statistics from the National Bureau of Statistics, the total fruit output reached 339.6581 million tons in 2024, indicating a broad market prospect. However, in the traditional field of mechanized harvesting, the structural limitations of rigid actuators severely restrict the efficiency of integrating robots with agriculture. Specifically, the isotropic gripping mode of traditional rigid grippers contradicts the anisotropic nature of the biomechanical properties of fruits: their planar fingertips struggle to match the complex curved surfaces of fruits, leading to uneven distribution of contact pressure and the formation of localized stress peaks. Studies have shown that these peaks can reach 1.8–2.3 times the yield strength of the epidermis, easily causing irreversible damage to fruit tissue.

[0003] To reduce harvesting damage, current research mainly focuses on compliant bionic hands and vacuum adsorption technology. These solutions achieve gentle harvesting to some extent through passive compliant or non-contact grasping via physical structures. For example, studies have shown that vacuum adsorption systems exhibit a lower probability of bruising after multiple impacts; dual-arm harvesting robots have also achieved a low overall damage rate, including the fruit loading process, in field tests. However, these existing technologies share a common, unresolved core deficiency: the lack of an active control mechanism to ensure the adaptive force field of multi-finger grasping. During the closing process, existing bionic dexterous hands, lacking real-time sensing and closed-loop control of the contact force direction, struggle to automatically converge the normal force vectors of each fingertip to the fruit's center of mass. This leads to instability in the grasping force field, generating slippage or additional compressive stress, which directly affects fruit harvesting quality and increases fruit damage rates. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for grasping quasi-circular objects based on dexterous hands. By collecting fingertip pressure data and initial posture information, and calculating the descriptive value and estimated center of mass reflecting the deviation of the applied force direction, the method achieves real-time and accurate perception of the contact force direction, solving the problem that existing technologies cannot perceive force vector deviation. At the same time, by dynamically adjusting the posture of each finger according to the initial posture and estimated center of mass, the method actively ensures that the applied force direction of all fingertips is accurately oriented towards the center of mass, fundamentally ensuring the self-organization and self-stability of the multi-finger grasping force field, effectively eliminating slippage and stress concentration caused by the divergence of the normal force vector. Finally, based on the formation of a stable grasping configuration, by preset a resultant force reference value and intelligently distributing the grasping pressure of each finger, and coordinating the tightening of the fingers, the method achieves accurate control of the overall clamping force within a safe range while ensuring a secure grasp. This proactively and effectively reduces the fruit damage rate at the control system level, improving the harvesting quality and reliability.

[0005] The technical solution of this invention is implemented as follows: On one hand, the present invention provides a method for grasping quasi-circular objects based on a dexterous hand, wherein the dexterous hand includes at least three fingers with a finger root rotation structure, and each fingertip is symmetrically provided with two pressure detection elements. The method for grasping quasi-circular objects based on a dexterous hand includes the following steps: S1 dexterous fingertip contacts the surface of a near-circular object and collects pressure detection data and initial pose information from each finger, the pressure detection data forming a pressure dataset; S2 obtains the description value corresponding to each fingertip based on the pressure dataset and obtains the estimated centroid of the circular object. The description value is used to reflect the deviation between the direction of the finger force and the normal of the surface of the circular object. S3 adjusts the position of each finger according to the initial position information, so that the fingertip of each finger faces the estimated center of mass of the quasi-circular object, forming a stable grasping configuration with force self-locking. S4 sets the resultant force reference value and allocates the gripping pressure parameters of each finger according to the resultant force reference value. The resultant force reference value is a preset overall gripping force for a round object. The S5 coordinates the tightening of all fingers, allowing the dexterous hand to grasp round objects based on the gripping pressure of each finger.

[0006] Based on the above technical solutions, preferably, obtaining the description value corresponding to each fingertip according to the pressure dataset includes the following steps: S21 retrieves data from two pressure sensors corresponding to a specific finger from the pressure dataset; S22 uses the data from the two pressure sensors as the first pressure value and the second pressure value, respectively. S23 calculates the pressure difference between the first pressure value and the second pressure value, and obtains the descriptive value based on the pressure difference.

[0007] More preferably, the formula for calculating the descriptive value D is: D = (P1 - P2) / (P1 + P2 + ε), where ε is a positive number, P1 is the first pressure value, and P2 is the second pressure value.

[0008] Based on the above technical solutions, preferably, in step S2, obtaining the estimated centroid of the circular object according to the pressure dataset includes the following steps: S24 For each finger, based on the corresponding description value and through a predefined mapping relationship, calculate the equivalent contact offset of the contact point in the local coordinate system of the fingertip; S25 obtains the spatial coordinates of the contact point based on the equivalent contact offset and the spatial position and attitude of the fingertip; S26 uses a spatial spherical fitting algorithm to fit the spatial coordinates of multiple contact points and solve for the optimal fitted sphere. S27 uses the coordinates of the center of the best-fit sphere as the estimated centroid of the circular object.

[0009] More preferably, the predefined mapping relationship is obtained through the following steps: The S241 drives a dexterous hand, enabling the fingertip of a single finger to press a calibration ball with known center coordinates in various different postures; S242 records the spatial position and posture of the fingertip, the pressure data of the two pressure sensors, and the actual offset of the actual contact point in the local coordinate system of the fingertip, calculated based on the geometric relationship between the fingertip posture and the center of the calibration ball, during each press. S243 uses a linear regression algorithm to fit the mapping coefficients based on the recorded multiple sets of descriptive values ​​and the actual offset.

[0010] Based on the above technical solution, preferably, step S3, adjusting the pose of each finger according to the initial pose information, includes the following steps: S31 combines the initial pose information of the finger with the estimated centroid coordinates to form an input feature vector; S32 inputs the input feature vector into the preset target pose prediction model, and outputs the target joint angle of the finger through the target pose model; S33 drives the finger movement according to the target joint angle, so that the fingertip is directed toward the estimated center of mass of the quasi-circular object.

[0011] Based on the above technical solutions, preferably, step S4 includes the following sub-steps: S41 preset resultant force reference value; S42 calculates the anti-slip weight factor for each finger based on a stable gripping configuration. The weight factor is positively correlated with the friction coefficient of the contact surface of the finger and the angle between the pressure direction and the gravity direction at the contact point. S43 distributes the resultant force reference value proportionally to each finger according to the anti-slip weighting factor, forming the gripping pressure parameters of each finger.

[0012] Based on the above technical solutions, preferably, step S6 is also included: during the grasping process, the pressure dataset is re-collected and the description value is updated to confirm whether the updated description value meets the formation conditions of a stable grasping configuration.

[0013] More preferably, step S6 includes the following sub-steps: S61 collects stress datasets during the grabbing process; S62 updates the description value based on the collected pressure dataset; S63 determines whether the description value of each finger is less than the preset description threshold; If S64 is true, then determine that the stable gripping configuration with self-locking force has been formed; otherwise, return to step S2.

[0014] On the other hand, the present invention provides a circular object grasping system based on a dexterous hand, which incorporates the above-mentioned circular object grasping method based on a dexterous hand.

[0015] The method and system for grasping quasi-circular objects based on dexterous hands of the present invention have the following advantages over the prior art: 1. By collecting fingertip pressure data and initial posture information, and calculating the descriptive value and estimated centroid reflecting the deviation of the applied force direction, real-time and accurate perception of the contact force direction is achieved, solving the problem that existing technologies cannot perceive force vector deviation. 2. By dynamically adjusting the position of each finger according to the initial position and the estimated center of mass, the force direction of all fingertips is precisely oriented towards the center of mass, which fundamentally ensures the self-organization and self-stability of the multi-finger grasping force field and effectively eliminates slippage and stress concentration caused by the divergence of the normal force vector. 3. Based on the formation of a stable gripping configuration, by preset the force reference value and intelligently distribute the gripping pressure of each finger, and coordinate the tightening of the fingers, it is possible to ensure a reliable grip while precisely controlling the overall clamping force within a safe range. This proactively and effectively reduces the fruit damage rate at the control system level, thereby improving the harvesting quality and reliability. Attached Figure Description

[0016] 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.

[0017] Figure 1 This is a schematic diagram of the process of the method for grasping circular objects based on dexterous hands according to the present invention.

[0018] Figure 2 This is a simulation diagram of the dexterous hand grasping method for grasping quasi-circular objects based on the present invention.

[0019] Figure 3 This is a simulation diagram of the resultant force analysis of the dexterous hand grasping method for grasping quasi-circular objects based on the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention based on the specific circumstances.

[0022] In the description of the embodiments of the present invention, it should be noted that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of the present invention.

[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0024] Embodiments of the present invention are described in detail below, examples of which are illustrated 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 intended to explain the present invention, and should not be construed as limiting the present invention.

[0025] The following disclosure provides numerous different embodiments or examples for implementing various structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. Additionally, examples of various specific processes and materials are provided, but those skilled in the art will recognize the applicability of other processes and / or the use of other materials.

[0026] like Figure 1-3 As shown, the present invention relates to a method for grasping quasi-circular objects based on a dexterous hand. The dexterous hand includes at least three fingers with a finger root rotation structure. Each finger has two pressure detection elements symmetrically arranged at its fingertip. The fingertip of the dexterous hand is flexible, so that its deformation can be transmitted to the pressure detection elements. At the same time, when grasping quasi-circular objects, the contact area can be increased by deformation, reducing local grasping pressure and avoiding damage during fruit picking. The two pressure detection elements can be strain gauges. It should be noted that the number of pressure detection elements can be three or more. The more pressure detection elements there are, the more pressure directions can be sensed, and the higher the accuracy will be. However, for the sake of explanation, this embodiment uses two pressure detection elements as an example.

[0027] The method for grasping circular objects based on dexterous hands includes steps S1-S6.

[0028] Step S1: The dexterous fingertip contacts the surface of a circular object and collects pressure detection data and initial pose information on each finger. The pressure detection data forms a pressure dataset.

[0029] This step is the initial stage of the grasping action. Its core purpose is to establish initial physical contact between the dexterous hand and the quasi-circular object, and simultaneously acquire the multimodal sensing data necessary for subsequent calculations. This step is the foundation for achieving subsequent adaptive grasping.

[0030] Specifically, the control unit drives the dexterous hand to move towards the target round object until the fingertips of at least three fingers make physical contact with the fruit surface. This stage employs a tentative, gentle contact strategy, controlling the fingers to close at a low speed and torque to ensure that the initial contact does not damage the fruit due to impact or excessive pressure. The goal of this initial contact is to form a preliminary, unstable envelope grasp, providing a basis for subsequent precise adjustments. This initial contact can be made after recognizing the approximate location of the target from an image.

[0031] Two pressure sensors symmetrically arranged at the fingertips of each finger detect the contact pressure at their location in real time. For a dexterous hand with n fingers, 2n independent pressure data points can be obtained in one acquisition. The set of these 2n pressure data points is the pressure dataset, which directly reflects the initial force on both sides of each fingertip.

[0032] By reading the position sensor readings on the rotational joints of each finger and other joints of the dexterous hand, such as high-precision encoders, and combining them with the kinematic model of the dexterous hand, the spatial position and posture of each fingertip at the moment of contact can be calculated in real time, i.e., the initial pose information.

[0033] Step S2: Based on the pressure dataset, obtain the description value corresponding to each fingertip and obtain the estimated centroid of the circular object. The description value is used to reflect the deviation between the direction of the finger force and the normal of the surface of the circular object.

[0034] In this embodiment, step S2 is a core calculation and state estimation step. Its purpose is to transform the original pressure dataset from step S1 into two high-level state variables with clear physical meaning and control value through a series of calculations: a descriptive value and a predicted centroid. This elevates the original, local contact information to a macroscopic judgment of the overall grasping stability, thereby providing precise targets and basis for subsequent proactive adjustments.

[0035] The descriptive value in this step is a key calculation parameter. Its core lies in quantifying the asymmetry of the pressure distribution of each fingertip and associating this asymmetry with the angular deviation of the finger's force direction relative to the normal of the object's surface. This is specifically achieved through steps S21-S23.

[0036] Step S21: Take the data from the two pressure sensors corresponding to a certain finger in the pressure dataset.

[0037] Step S21 is the data preparation stage for describing the value calculation. Its core lies in locating and extracting the raw sensing information related to a specific finger from the global pressure dataset. Specifically, the system accesses the pressure data collected and stored in step S1. This dataset contains readings of all pressure sensors on all fingers of the dexterous hand. For a specific finger to be calculated, the system accurately indexes and reads the instantaneous data of the two pressure sensors symmetrically installed at the fingertip at the same moment from this dataset, based on the predefined mapping relationship between sensor numbers and fingers.

[0038] Step S22: Use the data from the two pressure sensors as the first pressure value and the second pressure value, respectively.

[0039] The main task of this step is to identify and standardize the extracted raw data to facilitate clear mathematical operations and assign physical meaning. Specifically, two pressure data points extracted from a specified finger are assigned the labels P1 and P2, respectively. It should be noted that these labels are only used to distinguish two different sensing channels on the same fingertip during calculations. Whether they correspond to the left or right detection element must be uniformly agreed upon during system initialization and maintained consistently throughout the process. Through this step, two abstract voltage or digital signals are transformed into clearly directional physical quantities P1 and P2 that can be used for quantitative comparison and calculation.

[0040] Step S23: Calculate the pressure difference between the first pressure value and the second pressure value, and obtain the descriptive value based on the pressure difference.

[0041] The purpose of this step is to transform two independent pressure values ​​into a single and effective index that characterizes the degree of force balance at the fingertip. First, the original pressure difference between P1 and P2 is calculated. This difference directly reflects the absolute imbalance of pressure on both sides of the fingertip. In a specific embodiment, to further eliminate the influence of fluctuations in total pressure on the judgment result and improve the robustness of the system under different grasping forces, normalization is preferably used. The pressure difference is divided by the sum of the two pressure values, and a very small positive number ε is introduced to prevent the denominator from being zero. This yields the descriptive value D of the finger, a dimensionless quantity whose absolute value directly quantifies the degree of deviation between the force direction and the normal direction of the contact point. Its sign clearly indicates whether the deviation is biased towards the side containing P1 or P2, thus providing a precise and directional feedback signal for subsequent posture adjustments.

[0042] The specific formula for calculating the descriptive value D is: D = (P1 - P2) / (P1 + P2 + ε), where ε is a positive number, P1 is the first pressure value, and P2 is the second pressure value.

[0043] When the force applied by the fingertip is exactly along the surface normal of the contact point, the two pressure sensors are under equal pressure, P1≈P2, and the descriptive value D approaches zero. When the direction of the applied force deviates, causing one pressure sensor to be under greater pressure and the other under less pressure, P1≠P2, and the absolute value of the descriptive value D will increase. Its sign indicates the direction of the deviation. Therefore, the descriptive value D, as a dimensionless relative difference, effectively eliminates the influence of the absolute pressure magnitude and accurately reflects the degree and direction of deviation in the direction of the applied force.

[0044] To achieve more accurate geometric fitting, contact point coordinate correction is first required. Based on the initial fingertip pose recorded in step S1, the spatial coordinates of the fingertip's geometric center can be obtained. However, due to the volume of the fingertip, the actual contact point may not be at the geometric center. In this case, using the calculated descriptive value D and a predefined mapping relationship, the equivalent contact offset of the contact point in the fingertip's local coordinate system can be calculated. After converting this offset to the world coordinate system, it is superimposed with the fingertip center coordinates to obtain more accurate spatial coordinates of the contact point, specifically achieved through steps S24-S27.

[0045] Step S24: For each finger, calculate the equivalent contact offset of the contact point in the local coordinate system of the fingertip based on the corresponding description value and a predefined mapping relationship.

[0046] This step aims to solve the problem of the geometric center not coinciding with the actual contact point due to the physical size of the fingertip. Its core lies in converting the descriptive value, a mechanically sensed quantity, into a geometric offset in the local coordinate system of the fingertip. For each finger, the system calls the descriptive value D calculated in the previous step. This descriptive value is input into a predefined mapping relationship, which usually exists in the form of a function or lookup table. It describes the quantitative relationship between the descriptive value D and the equivalent contact offset Δ of the contact point along the sensing axis in the local coordinate system of the fingertip. For example, in a preferred embodiment, the mapping relationship is a linear function: Δ = k·D, where the mapping coefficient k is determined by prior calibration experiments. It should be noted that the equivalent contact offset is a vector whose direction is defined by the local coordinate system of the fingertip and whose magnitude is determined by the descriptive value D. The equivalent contact offset accurately characterizes the estimated position of the actual contact point relative to the geometric center of the fingertip.

[0047] Step S25: Obtain the spatial coordinates of the contact point based on the equivalent contact offset and the spatial position and orientation of the fingertip.

[0048] This step transforms the equivalent contact offset in the local coordinate system to the global world coordinate system, thereby obtaining the absolute spatial coordinates of the contact point. First, using the kinematic model of the dexterous hand and joint sensor readings, the spatial coordinates Q of the current fingertip geometric center and its attitude rotation matrix R are calculated. Then, the local offset Δ calculated in step S24 is transformed using the rotation matrix R to obtain its representation Δ_world in the world coordinate system. Finally, the corrected spatial coordinates Q' of the contact point, which are closer to the physical reality, are calculated using vector addition Q' = Q + Δ_world. This step effectively integrates the force-sensing-based local correction information with the robot's global spatial perception information, providing high-precision input data for subsequent geometric fitting.

[0049] Step S26: Use a spatial spherical fitting algorithm to fit the spatial coordinates of multiple contact points and solve for the optimal fitted sphere.

[0050] This step uses a spatial geometric fitting algorithm to reconstruct the surface morphology of the circular object, and uses the spatial coordinates of the contact points {Q'1, Q'2, ..., Q'} obtained in step S25 from all the fingers. n The point set is considered as a spatial point set located on the surface of a near-circular object. Subsequently, a spatial spherical fitting algorithm is used to process this point set. The goal of this algorithm is to find a spatial sphere that minimizes the sum of the squared distances from all contact points to this sphere. By solving this optimization problem, the system can calculate the equation of the optimally fitted sphere, which includes key parameters that determine the sphere's position and size.

[0051] In a specific embodiment, it is first necessary to establish a system of linear equations for a spatial point P. i (x i ,y i ,z i The equation of the sphere it satisfies can be transformed into the following linear form:

[0052] Where a, b, and c correspond to the three components to the left of the center of the optimally fitted sphere, and d is an intermediate parameter.

[0053] For all n contact points, a matrix equation is formed: A·X=B, where A is an n×4 matrix, and the elements in the i-th row are: [2x i ,2y i ,2z i [1], X is a 4×1 parameter vector to be determined: [a,b,c,d] T B is an n×1 vector, and its i-th element is: .

[0054] The parameter vector X is obtained by calculating the linear least squares solution:

[0055] Among them, A T It is the transpose of matrix A.

[0056] The optimal fit sphere has center coordinates (X0, Y0, Z0) corresponding to (a, b, c).

[0057] Step S27: Use the coordinates of the center of the best-fit sphere as the estimated centroid of the circular object.

[0058] From the optimal fitted spherical equation obtained in step S26, the coordinates of the sphere's center, O(X0, Y0, Z0), are directly read. For near-spherical objects such as apples and oranges, their geometric center and center of mass highly coincide. Therefore, using these sphere center coordinates as the estimated center of mass of the near-spherical object is a reasonable and effective approximation. These estimated center of mass coordinates will be directly used in subsequent pose adjustment steps as a common target for converging the force directions applied by the fingertips, thus providing a precise geometric reference for achieving a stable force-locking gripping configuration.

[0059] In addition, in some specific embodiments, the specific mapping relationship can be determined through steps S241-S243.

[0060] Step S241: Drive the dexterous hand to press a calibration ball with known center coordinates in a variety of different postures with the tip of a single finger.

[0061] A calibration ball with known precise center coordinates is fixed in the workspace. The control system drives a single finger of the dexterous hand to press the surface of the calibration ball with a series of pre-planned and differentiated spatial postures. These postures need to cover as many angle combinations as possible that the finger joint may occur in actual grasping, so as to ensure that the collected data has sufficient representativeness and broad coverage, laying the foundation for building a robust mapping relationship.

[0062] Step S242: In each press, record the spatial position and posture of the fingertip, the pressure data of the two pressure detection devices, and the actual offset of the actual contact point in the local coordinate system of the fingertip, calculated based on the geometric relationship between the fingertip posture and the center of the calibration ball.

[0063] The spatial position and attitude are obtained through the robot's kinematics model. The true offset of the actual contact point in the local coordinate system of the fingertip is the key truth label. The calculation is based on the known coordinates of the calibrated sphere center and the current geometric model of the fingertip. Through spatial geometric relationships, it is possible to accurately calculate which point on the fingertip surface is in contact with the sphere, and then obtain the component of the offset vector of the contact point relative to the geometric center of the fingertip in the direction of the sensing axis.

[0064] Step S243: Based on the recorded multiple sets of descriptive values ​​and the true offset, the mapping coefficients are fitted using a linear regression algorithm.

[0065] A linear regression algorithm is used to fit the dataset, and the mapping coefficient k in the linear model Δ=k·D is solved. This coefficient k is stored as the predefined mapping relationship, which is used to quickly and accurately estimate the equivalent contact offset Δ based on the real-time description value D during online crawling.

[0066] Step S3: Adjust the pose of each finger according to the initial pose information so that the fingertip of each finger is oriented toward the estimated center of mass of the quasi-circular object, forming a stable grasping configuration with force self-locking.

[0067] When the extensions of the normal forces applied by all fingers to the surface of the object intersect at the object's center of mass, the torques acting on the object in all directions reach equilibrium, resulting in the most stable grasping state, which achieves force self-locking. This step, through a one-time active posture adjustment, drives the fingertips of all fingers from their initial and potentially unstable contact postures to uniformly face and converge at the estimated center of mass, fundamentally avoiding slippage, squeezing, or rolling phenomena caused by force line divergence in traditional grasping.

[0068] In some embodiments, steps S31-S33 may be represented.

[0069] Step S31: Combine the initial pose information of the finger with the estimated centroid coordinates to form the input feature vector.

[0070] The initial pose information from step S1 is combined and spliced ​​with the estimated centroid coordinates O(X0,Y0,Z0) from step S2. The estimated centroid coordinates define the global target, while the initial joint angles fully describe the initial state of the dexterous hand at the moment of contact. These two parts of information together constitute an input feature vector that comprehensively describes the current system state and the target.

[0071] Step S32: Input the input feature vector into the preset target pose prediction model, and output the target joint angle of the finger through the target pose model.

[0072] The target pose prediction model is a machine learning model, such as a deep neural network, pre-trained with a large amount of data. Internally, it has learned the complex nonlinear mapping relationship between the initial state, the target centroid, and the optimal target joint angles. After receiving the input vector, the model directly outputs the target joint angles of all fingers through internal forward propagation calculations. This process replaces the complex inverse kinematics real-time solution and optimization process in traditional robot control, achieving efficient, direct, and intelligent pose planning.

[0073] Step S33: Based on the target joint angle, drive the finger movement so that the fingertip points toward the estimated center of mass of the quasi-circular object.

[0074] After obtaining the target joint angle vector, it is sent as a position command to the joint servo driver corresponding to each finger of the dexterous hand. The driver then controls the finger root joint motor to rotate, driving the finger to move smoothly from the initial pose to the target pose defined by the target joint angle in one go, so that its axis is accurately pointed to the estimated centroid of the quasi-circular object, thereby realizing the construction of a stable grasping configuration.

[0075] Step S4: Set the resultant force reference value and allocate the gripping pressure parameters of each finger according to the resultant force reference value. The resultant force reference value is a preset overall gripping force for a round object.

[0076] A global force reference value is set, which is based on the characteristics of a spherical object. Specifically, if applied to fruit picking, the type and ripeness of the fruit need to be considered. The upper limit of the safe gripping force is pre-set to ensure that the gripping force will not cause damage to the object from the system level. Secondly, based on physical principles and the current configuration, this total gripping force is intelligently decomposed and distributed to each finger, generating gripping pressure parameters that each finger needs to track independently. The force distribution can most effectively maintain gripping stability, thereby achieving a synergistic unity of overall safety and local stability.

[0077] In some specific embodiments, this is achieved through steps S41-S43.

[0078] Step S41: Preset the resultant force reference value.

[0079] A grasping force knowledge base can be set up. From the pre-stored grasping force knowledge base, based on the identification type of the round object to be grasped, such as the type of fruit such as apple or citrus, the corresponding resultant force reference value can be called. This value is usually set to a safe value that is slightly higher than the minimum force required to prevent the fruit from slipping, while being much lower than the critical force that would cause damage to its surface tissue. By presetting this value, the system sets a clear and safe global target for force control of the entire grasping process, making the grasping behavior controllable and precise.

[0080] Step S42: Based on the stable gripping configuration, calculate the anti-slip weighting factor for each finger. The weighting factor is positively correlated with the friction coefficient of the contact surface of the finger and the angle between the pressure direction and the gravity direction at the contact point.

[0081] This reflects the static friction characteristics between the fingertip material and the fruit peel. Contact points with a high coefficient of friction can withstand greater tangential friction, thus receiving a higher weight. The angle between the pressure direction and the gravity direction at the contact point determines how much of the normal force provided by the finger can be used to generate frictional force against gravity. The larger the angle, the higher the efficiency of the normal force against gravity, and the greater the required clamping force, thus receiving a higher weight. The weighting factor is positively correlated with these two factors, ensuring that the force distribution strategy prioritizes meeting the stability requirements for anti-slippage.

[0082] Step S43: Based on the anti-slip weighting factor, the resultant force reference value is proportionally allocated to each finger to form the gripping pressure parameters of each finger.

[0083] In this step, the preset resultant force reference value is proportionally distributed according to the anti-slip weighting factors of each finger calculated above. The specific calculation formula is: F_i=(W_i / ΣW)*F_ref, where F_i is the grasping pressure parameter of the i-th finger, ΣW is the sum of the weighting factors of all fingers, F_ref is the resultant force reference value, and W_i is the weighting factor. Through this calculation, the resultant force reference value is reasonably decomposed into a set of non-uniform force control commands acting on each finger. This set of commands ensures that the total grasping force remains constant within a safe range, while its distribution optimally matches the anti-slip requirements of each contact point under the current configuration, thus fundamentally and collaboratively solving the problem of firm grip and no damage.

[0084] Step S5: Coordinate the tightening of all fingers, and the dexterous hand grasps the round object according to the grasping pressure of each finger.

[0085] Each finger's joint actuator uses the grasping pressure parameter F_i obtained in step S43 as the set value of its force control loop, and reads the feedback data of the fingertip pressure detection device in real time. Specifically, the output torque of each finger joint can be dynamically adjusted through a high-frequency PID control algorithm or a more advanced impedance control algorithm, so that the actual pressure applied by each finger to the object surface quickly and accurately converges to its respective target value. This process is carried out by all fingers in coordination. The coordination between the fingers is reflected in the synchronous tracking of the unified force distribution scheme, rather than simple synchronous movement, thereby ensuring that while forming a stable grip, the overall grasping force is precisely constrained within the preset safety range, and finally a high-quality grasping with both robustness and compliance is completed.

[0086] Step S6: During the crawling process, re-collect the stress dataset and update the description value, and confirm whether the updated description value meets the formation conditions of a stable crawling configuration.

[0087] This step, by establishing a real-time feedback channel, effectively addresses changes in the grasping state caused by fruit plastic deformation, external disturbances, or initial estimation errors, ensuring the system can maintain its optimal force-locking configuration over the long term. Its technical value lies in transforming static grasping into a dynamically adaptive and stable process, effectively preventing gradual degradation of the grasping state through periodic verification, and providing the ultimate guarantee for highly reliable grasping. Specifically, this step implements a complete verification loop from data acquisition to decision execution through four sub-steps.

[0088] Step S61: Collect stress dataset during the capture process.

[0089] While maintaining the current grasping state, readings from all finger pressure sensors are synchronously acquired at a specific sampling period, reconstructing a real-time pressure dataset containing dual-channel pressure values ​​from each fingertip. Unlike the initial contact acquisition, this step involves data acquisition under the influence of grasping force, and its data characteristics simultaneously reflect both object deformation properties and stable contact state, providing more timely perceptual input for state assessment. This continuous data stream provides the system with the perceptual basis for monitoring changes in the grasping state.

[0090] Step S62: Update the description value based on the collected pressure dataset.

[0091] Following the same algorithm as step S2, based on the real-time pressure dataset collected in step S61, an updated descriptive value is independently calculated for each finger. The updated descriptive value reflects the balance of force distribution at each fingertip in the current grasping state and is a direct indicator of whether the force lines still converge at the center of mass. Compared with the initial descriptive value, the updated descriptive value better reflects the force distribution characteristics during the stable grasping phase, providing accurate feature input for stability judgment.

[0092] Step S63: Determine whether the description value of each finger is less than the preset description threshold.

[0093] In this step, the updated description value of each finger needs to be compared with a preset description threshold. This threshold is an empirical value determined based on a large number of experiments, and is usually set as a positive decimal close to zero. The criterion is that the stability condition is met only when the absolute value of the description value of all fingers is less than the threshold.

[0094] Step S64: If yes, determine that the force-locking stable gripping configuration has been formed; otherwise, return to step S2.

[0095] If the judgment result of step S63 is that the stability condition is met, it confirms that the stable grasping configuration of the self-locking force has been formed or maintained, and the grasping task can continue to be executed or completed; if the description value of any finger exceeds the threshold range, it indicates that the current grasping state has deviated from the optimal configuration, and the system will automatically return to step S2 and restart the complete process from centroid estimation to pose adjustment.

[0096] On the other hand, the present invention provides a circular object grasping system based on a dexterous hand, which incorporates the above-mentioned circular object grasping method based on a dexterous hand.

[0097] Specifically, this system, through deep integration of hardware architecture and software algorithms, materializes each step in the aforementioned method embodiments into dedicated functional modules. The system includes: a multi-finger dexterous hand body containing at least three bionic fingers with finger root rotation structures, each finger tip symmetrically integrating two high-precision pressure detection elements; a multi-axis motion control module for driving each finger joint and acquiring joint pose information in real time; a data acquisition and processing module connected to the pressure detection elements for synchronously acquiring pressure datasets and performing signal conditioning; a central processing unit embedded with a control algorithm programmed according to any of the aforementioned method embodiments, specifically for performing descriptive value calculation, centroid estimation, pose planning, and force distribution decisions; and a closed-loop control module for collaboratively controlling the force / position movement of each finger according to instructions from the central processing unit. Through the collaborative operation of these modules, the system achieves full automation from initial contact perception to stable force-controlled grasping, providing a complete hardware solution for non-destructive grasping of spherical objects.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for grasping quasi-circular objects based on a dexterous hand, characterized in that, The dexterous hand includes at least three fingers with a root rotation structure, and each finger has two pressure detection elements symmetrically arranged at its tip. The dexterous hand-based method for grasping quasi-circular objects includes the following steps: S1 dexterous fingertip contacts the surface of a near-circular object and collects pressure detection data and initial pose information from each finger, the pressure detection data forming a pressure dataset; S2 obtains the description value corresponding to each fingertip based on the pressure dataset and obtains the estimated centroid of the circular object. The description value is used to reflect the deviation between the direction of the finger force and the normal of the surface of the circular object. S3 adjusts the position of each finger according to the initial position information, so that the fingertip of each finger faces the estimated center of mass of the quasi-circular object, forming a stable grasping configuration with force self-locking. S4 sets the resultant force reference value and allocates the gripping pressure parameters of each finger according to the resultant force reference value. The resultant force reference value is a preset overall gripping force for a round object. The S5 coordinates the tightening of all fingers, allowing the dexterous hand to grasp round objects based on the gripping pressure of each finger.

2. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 1, characterized in that, The step of obtaining the description value corresponding to each fingertip based on the pressure dataset includes the following steps: S21 retrieves data from two pressure sensors corresponding to a specific finger from the pressure dataset; S22 uses the data from the two pressure sensors as the first pressure value and the second pressure value, respectively. S23 calculates the pressure difference between the first pressure value and the second pressure value, and obtains the descriptive value based on the pressure difference.

3. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 2, characterized in that, The formula for calculating the descriptive value D is: D = (P1 - P2) / (P1 + P2 + ε), where ε is a positive number, P1 is the first pressure value, and P2 is the second pressure value.

4. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 1, characterized in that, In step S2, the predicted centroid of the circular object is obtained based on the pressure dataset, including the following steps: S24 For each finger, based on the corresponding description value and through a predefined mapping relationship, calculate the equivalent contact offset of the contact point in the local coordinate system of the fingertip; S25 obtains the spatial coordinates of the contact point based on the equivalent contact offset and the spatial position and attitude of the fingertip; S26 uses a spatial spherical fitting algorithm to fit the spatial coordinates of multiple contact points and solve for the optimal fitted sphere. S27 uses the coordinates of the center of the best-fit sphere as the estimated centroid of the circular object.

5. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 4, characterized in that, The predefined mapping relationship is obtained through the following steps: The S241 drives a dexterous hand, enabling the fingertip of a single finger to press a calibration ball with known center coordinates in various different postures; S242 records the spatial position and posture of the fingertip, the pressure data of the two pressure sensors, and the actual offset of the actual contact point in the local coordinate system of the fingertip, calculated based on the geometric relationship between the fingertip posture and the center of the calibration ball, during each press. S243 uses a linear regression algorithm to fit the mapping coefficients based on the recorded multiple sets of descriptive values ​​and the actual offset.

6. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 1, characterized in that, In step S3, the poses of each finger are adjusted according to the initial pose information, including the following steps: S31 combines the initial pose information of the finger with the estimated centroid coordinates to form an input feature vector; S32 inputs the input feature vector into the preset target pose prediction model, and outputs the target joint angle of the finger through the target pose model; S33 drives the finger movement according to the target joint angle, so that the fingertip is directed toward the estimated center of mass of the quasi-circular object.

7. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 1, characterized in that, Step S4 includes the following sub-steps: S41 preset resultant force reference value; S42 calculates the anti-slip weight factor for each finger based on a stable gripping configuration. The weight factor is positively correlated with the friction coefficient of the contact surface of the finger and the angle between the pressure direction and the gravity direction at the contact point. S43 distributes the resultant force reference value proportionally to each finger according to the anti-slip weighting factor, forming the gripping pressure parameters of each finger.

8. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 1, characterized in that, It also includes step S6, in which the stress dataset is re-collected during the grabbing process, the description value is updated, and it is confirmed whether the updated description value meets the formation conditions of a stable grabbing configuration.

9. The method for grasping quasi-circular objects based on a dexterous hand as described in claim 8, characterized in that, Step S6 includes the following sub-steps: S61 collects stress datasets during the grabbing process; S62 updates the description value based on the collected pressure dataset; S63 determines whether the description value of each finger is less than the preset description threshold; If S64 is true, then determine that the stable gripping configuration with self-locking force has been formed; otherwise, return to step S2.

10. A circular object grasping system based on a dexterous hand, characterized in that, The method for grasping circular objects based on a dexterous hand, as described in any one of claims 1-9, is embedded.

Images