A dexterous hand fruit and vegetable grabbing control method based on fuzzy variable impedance control

Abstract

The application discloses a dexterous hand fruit and vegetable grabbing control method based on fuzzy variable impedance control. First, a multi-finger cooperative kinematics model is established based on the screw theory to determine the fingertip pose, the grabbing force of each finger and the multi-finger force distribution; an optimization model with the minimum contact force as the target is constructed based on force closure, and the NSGA-II algorithm is used for solving to obtain the minimum grabbing force required for stable and lossless grabbing; an impedance control model is established in the Cartesian space, the grabbing system is equivalent to a mass-spring-damper system, the influence of the impedance parameters on the grabbing performance is analyzed, and the inertia parameter and the damping coefficient are determined as the key adjustment objects; further, fuzzy control is introduced, the grabbing force error and its change rate are taken as the inputs, the inertia and the damping adjustment amount are taken as the outputs, a fuzzy adaptive variable impedance controller is constructed, and dynamic optimization of the impedance parameters and accurate tracking of the grabbing force are realized. The application can be effectively applied to the fast, stable and lossless grabbing of various fruits and vegetables in unstructured agricultural environments.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to the non-destructive grasping and control of fruits and vegetables by a dexterous hand in an agricultural harvesting robot. Background Technology

[0002] With the continuous development of intelligent agricultural equipment, automated and intelligent fruit and vegetable harvesting robots have gradually become a research hotspot in the agricultural harvesting field. As a key component that directly interacts with fruits and vegetables, the performance of the end effector directly affects harvesting efficiency and fruit quality. Currently, dedicated actuators designed for specific fruits and vegetables have poor adaptability when dealing with fruits of different shapes, hardness, and surface characteristics. In contrast, multi-fingered dexterous hands, due to their flexible structure and diverse grasping modes, exhibit greater versatility and adaptability, and are expected to become the ideal actuator for fruit and vegetable harvesting robots.

[0003] However, dexterous hands still face many challenges in fruit and vegetable harvesting applications. Their multi-finger, multi-joint nature leads to complex grasping plans; factors such as grasping posture, contact point selection, and force distribution all affect the grasping effect. More importantly, there is a contradiction between "stable grip" and "damage": excessive grasping force can cause mechanical damage to fruits and vegetables, while insufficient force can easily cause them to slip. Furthermore, fruits and vegetables are inherently fragile, easily deformable, and have unknown surface properties. Combined with interference factors in the unstructured agricultural environment, such as vibration, shading, and changes in lighting, this further increases the difficulty of dexterous hand grasping control.

[0004] In terms of control strategies, traditional impedance control can adjust the grasping force by simulating the behavior of a mass-spring-damped system. However, its parameters are usually fixed, making it difficult to adapt to the different physical characteristics of fruits and vegetables and dynamic environmental changes. Fixed parameters can easily lead to overshoot or lag in force response, resulting in fruit damage or harvesting failure. Therefore, there is an urgent need for a control method that can dynamically adjust impedance parameters and adapt to various grasping conditions to enable dexterous hands to grasp fruits and vegetables quickly, stably, and without damage. Summary of the Invention

[0005] To achieve the above objectives, this invention proposes a dexterous hand grasping control method based on fuzzy variable impedance control, comprising the following technical steps to solve the technical problem:

[0006] Step 1: Construct a kinematic model of dexterous hand grasping fruits and vegetables, establish a multi-finger cooperative kinematic model based on spinor theory, and determine the pose of each fingertip in the base coordinate system, the grasping force of each finger, and the distribution of the resultant force of multiple fingers;

[0007] Step 2: Based on force closure, construct a multi-objective gripping force optimization model with minimum contact force and minimum deviation between contact force and normal direction as the criterion. Solve the model using the NSGA-II algorithm to obtain the minimum gripping force required for stable and non-destructive gripping.

[0008] Step 3: Establish an impedance control model in Cartesian space, and then simulate the established impedance control model to obtain candidate values ​​for impedance control parameters, including candidate inertia coefficients. Candidate damping coefficient Candidate stiffness coefficient ;

[0009] Step 4: Introduce fuzzy control, using the grasping force tracking error and its rate of change as input, and outputting the inertia adjustment ΔM and damping adjustment ΔB, to construct a fuzzy adaptive variable impedance controller based on candidate inertia coefficients. and candidate damping coefficient Calculate the inertia coefficient of the fuzzy adaptive variable impedance controller and damping coefficient The stiffness coefficient K of the fuzzy adaptive variable impedance controller will be set as the candidate stiffness coefficient. Fuzzy adaptive variable impedance controller based on inertia coefficient and damping coefficient It uses the desired pose, velocity, and acceleration of the fingertip in Cartesian space coordinates, the actual pose, velocity, and acceleration of the fingertip in Cartesian space coordinates, and the set stiffness coefficient K and desired gripping force as inputs to achieve precise tracking of the gripping force.

[0010] Furthermore, in step 1, a single-finger kinematic model is established using spinor theory, including: establishing the transformation relationship between the base coordinate system of each finger and the global coordinate system; using the exponential product formula to represent the fingertip pose; calculating the spatial Jacobian matrix of each finger; and establishing the mapping relationship between joint torque and contact force.

[0011] Furthermore, in step 2, the gripping force optimization model includes the following constraints: force balance constraint, friction cone constraint, and joint torque constraint.

[0012] Furthermore, in step 3, the impedance control model is:

[0013]

[0014] In the formula, , , These are the desired inertia coefficient matrix, the desired damping coefficient matrix, and the desired stiffness coefficient moment, respectively. This represents the actual pose of the robotic arm's end effector. The desired pose of the robotic arm's end effector. For actual grasping power, For the desired grasping force.

[0015] Furthermore, the candidate inertia coefficient and candidate damping coefficient The desired stiffness coefficient matrix K represents the key adjustment parameters determined through simulation analysis. , .

[0016] Furthermore, in step 4, fuzzy control is introduced, taking the grasping force tracking error and its rate of change as input, and outputting the inertia adjustment ΔM and damping adjustment ΔB, including:

[0017] Grasping force error and error change rate The linguistic variables are fuzzed using the triangular membership function: {NB, NM, NS, ZO, PS, PM, PB}.

[0018] Based on the preset fuzzy rule table, fuzzy reasoning is performed to obtain the fuzzy outputs of the inertia change ΔM and the damping change ΔB.

[0019] The fuzzy rule is: when the grasping force error e f Belongs to fuzzy sets And the rate of change of error Belongs to fuzzy set B h When the change in the output inertia coefficient ΔM belongs to the fuzzy set... Meanwhile, the change in damping coefficient ΔB belongs to a fuzzy set. ;

[0020] Fuzzy set in the fuzzy rule table of inertia coefficient change The row corresponding to the middle subset and the fuzzy set B h The membership degree of a subset whose columns are orthogonal is a fuzzy set. A subset of;

[0021] Fuzzy set in the fuzzy rule table of damping coefficient change The row corresponding to the middle subset and the fuzzy set B h The membership degree of a subset whose columns are orthogonal is a fuzzy set. A subset of.

[0022] Furthermore, the dexterous hand control system involved in the method includes:

[0023] The kinematic modeling module is used to construct dexterous hand grasping models based on spinor theory.

[0024] An optimized calculation module is used to solve for the minimum gripping force using the NSGA-II algorithm;

[0025] Impedance control module is used to achieve initial tracking of the gripping force;

[0026] The fuzzy inference module is used to dynamically adjust the impedance parameters M and B based on the grasping force error and the rate of change.

[0027] Compared with the prior art, the present invention has the following advantages:

[0028] 1. A balance between gripping stability and fruit non-damage is achieved: By constructing an optimization model based on force closure theory and using the NSGA-II algorithm to solve for the minimum necessary gripping force, the gripping force can be minimized while ensuring gripping stability and preventing fruits and vegetables from slipping, thereby effectively avoiding mechanical damage to fruits and vegetables.

[0029] 2. Significantly improved system adaptability in unstructured environments: By introducing fuzzy control logic, a fuzzy adaptive variable impedance controller was constructed, which can adjust the impedance parameters in real time and dynamically according to the grasping force error and its rate of change, overcoming the shortcomings of traditional impedance control parameters being fixed and unable to cope with the differences in physical characteristics of fruits and vegetables and environmental interference.

[0030] 3. Optimized dynamic response performance during the grasping process: Through simulation analysis of impedance parameters and design of fuzzy rules, the system can intelligently adjust its response characteristics according to actual conditions. For example, it responds quickly when the error is large and suppresses overshoot and oscillation when approaching the target, thereby achieving faster response speed, smaller overshoot, and higher steady-state accuracy.

[0031] 4. Solved a key contradiction in agricultural harvesting: This method directly addresses the core contradiction between "firm grip" and "no damage." Through a comprehensive solution that integrates kinematics, optimization algorithms, and intelligent control, it provides an effective technical path for dexterous hands to reliably grasp fruit in complex and ever-changing unstructured agricultural environments (such as vibration, shading, and fruit diversity).

[0032] 5. Improved intelligence and automation of control: This method does not rely on precise fruit and vegetable models and environmental parameters. It can automatically complete online optimization of parameters and precise tracking of grasping force through sensor feedback and fuzzy inference, reducing the system's dependence on prior knowledge and enhancing the robustness and practicality of the entire harvesting system. Attached Figure Description

[0033] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 For dexterity hand coordinates;

[0035] Figure 2 A schematic diagram of the axis and coordinates of the thumb joint in a dexterous hand;

[0036] Figure 3 This is a transformation from the contact coordinate system to the fruit coordinate system;

[0037] Figure 4 For impedance parameter simulation experiments;

[0038] Figure 5 Here is a block diagram of a fuzzy variable impedance controller system;

[0039] Figure 6 For a fuzzy language system;

[0040] Figure 7 This is a schematic diagram of point contact.

[0041] Figure 8 This is a non-destructive contact model;

[0042] Figure 9 This is a test bench for measuring the coefficient of friction.

[0043] Figure 10 This is a test bench for non-destructive gripping force measurement.

[0044] Figure 11 This is a test bench for non-destructive operating force measurement;

[0045] Figure 12 For the conical part of the contact model;

[0046] Figure 13 This is a three-dimensional response diagram. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer and to facilitate understanding and implementation by those skilled in the art, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0048] A dexterous hand grasping control method based on fuzzy variable impedance control includes the following steps:

[0049] Step 1: Construct a dexterous hand grasping fruit and vegetable model. Based on spinor theory, establish a kinematic model of the dexterous hand through the base coordinate relationship of the fingers. Calculate the flexion / extension angle and adduction / abduction angle of the corresponding joints based on the motion information of each finger joint. Transform the kinematic model module into the actual fingertip Cartesian position to obtain the fingertip pose of a single finger when the dexterous hand grasps fruit and vegetables. Measure the actual fingertip grasping force using a thin-film force sensor at the fingertip of the dexterous hand to obtain the resultant grasping force of multiple fingers when grasping fruit and vegetables.

[0050] like Figure 1 Establish a global coordinate system at the base joint of the middle finger. The coordinates of the base joints of the thumb, index finger, middle finger, ring finger, and little finger are represented as follows: , .

[0051] The pose of each finger's base joint relative to the middle finger's base joint is as follows: (1)

[0052] In the formula, Let be the transformation matrix of the base coordinates of the i-th finger relative to the base joint of the middle finger. The coordinates are for the base joint of the middle finger.

[0053] Each finger of a dexterous hand can be represented as an equivalent robotic arm, thus decomposing the dexterous hand into multiple sets of robotic arms to obtain the kinematic model of the whole hand. A coordinate system for each finger is established, and the fingertip position and Jacobian matrix of each finger are solved.

[0054] definition Let be the coordinate transformation matrix from the base coordinate system to the end-effector coordinate system when the robotic arm is in the zero position. Here are the direction vectors of each joint axis. These are the position vectors of each joint axis. Let j be the rotational motion of the j-th joint. , It is the angle of the j-th joint. It is the motion spinor matrix in the form of the "hat operator" of the j-th joint.

[0055] Based on the POE kinematic model of an n-DOF manipulator, the fingertip pose of the j-th joint of the i-th finger in its base joint coordinate system is generated. (2)

[0056] in, , —— The corresponding special orthogonal group Lie algebra, .

[0057] We'll analyze the thumb; the other fingers are analyzed using the same method. For example... Figure 2As shown, establish a thumb-based coordinate system. With fingertip coordinate system .

[0058] The kinematic parameters of the thumb POE model are shown in Table 1. In the zero position state, the fingertip coordinate system... Relative thumb base coordinate system The transition matrix is (3)

[0059] Table 1. Kinematic parameters of the exponential product of thumb.

[0060]

[0061] Therefore, substituting the parameters from Table 1 and Equation 3 into Equation 2, we obtain the thumb tip posture of the dexterous hand as follows: (4)

[0062] In the formula, , , . express , express , express , express r and e are taken as 1 and 2, respectively. — Thumb The positional variables of the joints, r and e, are both less than the maximum value of the number of finger joints j;

[0063] For the spatial Jacobian of a serial robot, spinor theory has... (5)

[0064] in Ad represents the adjoint transformation.

[0065] Substituting the model parameters from Table 1 into Equation 5, we obtain the spatial Jacobian of the thumb as follows: (6)

[0066] Joint torque in contact coordinate system The mapping relationship with contact force is as follows: (7)

[0067] The instantaneous spatial velocity of the thumb tip can be expressed as a kinetic spinor: (8)

[0068] In the formula, Let be the configuration space of the thumb tip in the thumb base coordinate system.

[0069] Based on the fruit and vegetable grasping model, establish a contact coordinate system, a fruit and vegetable centroid coordinate system, and a world coordinate system. Select the fruit and vegetable centroid coordinate system as the world coordinate system. Figure 3 As shown.

[0070] The i-th contact coordinate system Relative to the fruit and vegetable coordinate system The position is The contact force is expressed in the fruit and vegetable coordinate system as: (9)

[0071] in, In the formula —The grasping matrix for the i-th contact point, —A force helical base with frictional contact points, —A 3×3 identity matrix;

[0072] When k fingers exert contact forces on a fruit, the mapping relationship between the contact force spiral and the external force spiral is as follows: (10)

[0073] in, ,

[0074] In the formula, To capture the matrix; The contact force matrix; The external force spiral that is balanced by the resultant force of contact.

[0075] Step 2: Determine the stable picking judgment rule based on force closure, constrain the picking force balance, non-destructive contact linearization, and joint torque, construct a grasping force optimization model for grasping fruits and vegetables, and optimize the minimum contact force based on the NSGA-II algorithm to obtain the minimum grasping force required for grasping fruits and vegetables, i.e., obtain the desired grasping force. The goal is to achieve stable grasping with the minimum grasping force and operating force, ensuring that fruits and vegetables are not damaged.

[0076] A minimum gripping force index function is established, using the minimum gripping force at each contact point as the index. , The smaller the value, the higher the rate of fruit and vegetable damage-free processing. (11)

[0077] in,

[0078] In the formula, For gripping force matrix; This is the grasping force weight matrix. Since dexterous hand grasps by using finger-to-finger contact, the finger-to-finger contact area is divided into k points of contact. The force at each point in the z-axis direction, such as Figure 7 As shown.

[0079] Friction cone angle at the contact point The angle between the contact force and the normal at the i-th contact point ,in , , This refers to the force at each point along the x, y, and z axes. The goal is to minimize the gripping force at each fingertip while ensuring the fingertip force is closest to the normal direction, thus establishing a minimum operational force index function. , (12).

[0080] It is worth noting that the friction cone angle It is unknown; the coefficient of friction between the finger and the fruit needs to be calculated first. This solution process requires the use of instruments. The specific method is as follows:

[0081] Dexterous hands primarily rely on manipulating force for harvesting; therefore, it is assumed that there is frictional contact between the fingers and the fruit and vegetable skins, and the rolling or sliding of the fingers on the fruit and vegetable skins is not considered. Figure 8 As shown, the contact force when grasping fruits and vegetables can be expressed as: , , , Contact force In the x, y, and z directions, the contact force is located inside the friction cone, meaning the operating force is less than or equal to the gripping force and the coefficient of friction. The product of the contact force and the gripping force perpendicular to the contact surface is the main cause of damage to fruits and vegetables, while the operational force tangential to the contact surface can cause abrasions to the peel. Simultaneously, contact forces closer to the edge of the friction cone increase the likelihood of slippage. Therefore, the contact force on fruits and vegetables should be kept within a certain threshold to avoid uncertain damage, improve contact stability, and decompose the maximum contact force under non-damaging conditions into non-damaging gripping forces. and non-destructive operating force Therefore, the constraints of the frictional non-destructive point contact model are:

[0082] (13)

[0083] like Figure 9As shown, peaches are more susceptible to damage than other fruits and vegetables. Therefore, the "Xiahui No. 5" peach variety was used as the test object to determine its static friction coefficient. The measuring equipment was an MXD-02 friction coefficient meter (Jinan Langguang Electromechanical Technology Co., Ltd.), and the contact material was polyurethane rubber with a Shore hardness of 40A. During the test, the polyurethane rubber (150mm×80mm×2mm) was fixed to the test slide using a clamp. A tension sensor and the peach were connected by a thin rope, and the peach was placed stably on the contact material. The mass of the weights on the friction coefficient meter was set to be equal to the mass of the peach, and the moving speed of the test slide was 100mm / min. To reduce the experimental error, three peach samples were tested three times each, and the average value was taken. The test results are shown in Table 2.

[0084] Table 2 Results of Static Friction Coefficient Measurement

[0085]

[0086] Among these, the non-destructive contact force can be decomposed into a non-destructive gripping force. and non-destructive operating force Lossless gripping power The testing platform mainly includes a dexterous hand and an FSR402 type thin-film pressure sensor, such as... Figure 10 As shown. A thin-film pressure sensor is placed under the fingertip material. A nimble fingertip applies a stable pressure towards the center of mass of the peach directly above it for 10 seconds, and the value of the thin-film pressure sensor is recorded, i.e., the gripping force. Gripping force tests were conducted in 0.5N increments from 2 to 8N, for a total of 13 groups. After being placed at room temperature (25℃) for 48 hours, the browning of the peach skin and flesh was observed. The maximum gripping force in the group with no obvious browning was recorded as the non-destructive gripping force. .

[0087] Non-destructive operating force The testing platform mainly includes an iron frame, a dexterous hand, an FSR402 type thin-film pressure sensor, and a digital force gauge (0~50.00N, accuracy: 0.01N), such as... Figure 11 As shown. A peach is placed on a tray, which is vertically connected to an iron frame by a thin rope, while the peach is horizontally connected to a digital tension gauge. Dexterous fingertips apply steady pressure directly above the peach, maintaining the membrane pressure sensor at a specified value. The operating force was defined as the altered reading of a digital tension meter, which was gradually increased and held for 10 seconds. The operating force ranged from 2 to 8 N, with 13 groups tested at 0.5 N intervals. The peaches were placed at 25°C for 48 hours, and the browning of the skin and flesh was observed. The maximum operating force in the group with no obvious browning was recorded as the non-destructive operating force.

[0088] The mechanical properties of the peaches measured in the experiment were as follows: the average static friction coefficient was 0.51, the average non-destructive gripping force was 7.5N, and the average non-destructive operating force was 3.5N.

[0089] The upper semi-conical part of the non-destructive grasping model can be disassembled into Figure 12 The normal is a line perpendicular to the plane (in Figure 12 The visible normal is the z-axis, and α is the angle between the cone and the normal.

[0090] The range of values ​​for is ( ,when At this point, the contact force is located on the conical surface of the friction cone, at the critical point of relative sliding; when At that time, the contact force coincides with the normal of the cone, resulting in the most stable grip.

[0091] To minimize the force, construct an optimization objective function. The multi-objective contact force optimization model for non-destructive harvesting of fruits and vegetables can be summarized as follows:

[0092]

[0093] In the formula, Indicates the decision quantity. ; For harvesting power; The external force spiral that is balanced by the resultant force of the contact; , These are the maximum and minimum joint torques, respectively. Constraints for the frictional non-destructive point contact model.

[0094] Optimization of picking contact force based on NSGA-II algorithm to obtain the desired grasping force.

[0095] Step 3: Based on dexterous hand dynamics and kinematics, determine the relationship matrix from the dexterous fingertip task space to the joint space. Treat the dexterous hand grasping fruits and vegetables as a second-order system of a virtual mass-spring-damping model, and establish an impedance control model in the Cartesian coordinate system.

[0096] The dynamic equations of the fingers of a dexterous hand in joint space are as follows:

[0097] (15)

[0098] In the formula, The inertia matrix, For Coriolis matrix, The gravitational torque vector, and These are the joint torque and external torque vectors, respectively.

[0099] From the kinematic model of the dexterous hand in step 1, the actual pose, velocity, and acceleration of the fingertip in Cartesian coordinates are obtained as follows: (16)

[0100] (17)

[0101] (18)

[0102] External torque With generalized external force vector Joint torque With fingertip torque The relationships are as follows:

[0103] (19)

[0104] (20)

[0105] By rearranging equations (15) to (20), we can obtain the dynamic model of a dexterous fingertip in Cartesian space as follows:

[0106] (twenty one)

[0107] The matrix from task space to joint space has the following relationship:

[0108] (twenty two)

[0109] The impedance control model in Cartesian coordinates is as follows:

[0110] (twenty three)

[0111] In the formula, , , These are the expected inertia coefficient matrix, expected damping coefficient matrix, and expected stiffness coefficient matrix of the impedance model, respectively. This refers to the actual position of the fingertips. The desired pose of the fingertip. For actual grasping power, For the desired grasping power, To account for the gripping force error, .

[0112] Actual gripping force The actual pose of the fingertip is measured by a thin-film force sensor. The control rate is obtained through the position sensor:

[0113] (twenty four)

[0114] The fruit and vegetable model is simplified into a spring system:

[0115] (25)

[0116] In the formula, For the local normal stiffness of fruits and vegetables, This refers to the position of the fruit and vegetable peel when there is no contact.

[0117] Obtain the desired pose of the fingertip based on the desired grasping force. (26)

[0118] Step 4: Set the impedance control parameters in the impedance control model obtained in Step 3, and then simulate the established impedance control model, including: setting one of the impedance control parameters, the desired inertia coefficient matrix, the desired damping coefficient matrix, and the desired stiffness coefficient matrix to n different values, and setting the other two to constant values, to obtain the impedance control model corresponding to each of the n sets of impedance control parameters under three simulation conditions, and using the actual fingertip pose, the desired fingertip pose, and the desired gripping force as inputs to the impedance control model to obtain the actual gripping force output by the impedance control model.

[0119] The impedance control models corresponding to each of the n sets of impedance control parameters under the three simulation conditions include:

[0120] The first method involves simulating the stiffness coefficient K, setting K to vary from 500 N / m to 1000 N / m, and setting M = 1 N·s. 2 / m, B=20N·s / m, the local normal stiffness of fruits and vegetables is 1000N / m, and the expected gripping force is 5N.

[0121] The second method involves simulating the inertia coefficient M, setting M to a value starting from 1 N·s. 2 / m changes to 5N·s 2 / m, and set B=20N·s / m, K=1000N / m, fruit and vegetable stiffness is 1000N / m, and expected gripping force is 5N.

[0122] The third method involves simulating the damping coefficient B, setting B to vary from 8 N·s / m to 50 N·s / m, and setting M = 0.1 N·s. 2 / m, K=1000N / m, the stiffness of fruits and vegetables is 1000N / m, and the expected gripping force is 5N.

[0123] It is worth noting that the stiffness coefficient K of fruits and vegetables is not the same as the local normal stiffness of fruits and vegetables. However, both must be set to the same size.

[0124] Simulation results are as follows Figure 4 .

[0125] The stiffness coefficient is negatively correlated with the steady-state error of the gripping force. When the stiffness coefficient is larger, the actual position of the fingertip is closer to the desired position. Increasing the stiffness coefficient improves the finger's ability to approach the desired position and desired gripping force, resulting in a smaller steady-state deviation.

[0126] A larger inertia coefficient results in a slower increase in gripping force and a longer oscillation period. A smaller inertia coefficient allows the gripping force to respond more quickly to step commands, shortening the transition time and making the system more stable.

[0127] An excessively large damping coefficient sacrifices the gripping force response speed, while an excessively small damping coefficient makes it difficult to stabilize the gripping force oscillation.

[0128] The stiffness coefficient K primarily affects the system's steady-state error, with a relatively small impact on dynamic performance. A larger K helps reduce the pressing depth and gripping force, thereby lowering the risk of damage to fruits and vegetables. Therefore, K is set to a large fixed value (without dynamic adjustment). The inertia coefficient M and damping coefficient B significantly affect the system's dynamic response performance; therefore, the impedance control parameter affecting the stability of the gripping force is mainly the inertia coefficient. and damping coefficient Candidate values ​​for impedance control parameters, including candidate inertia coefficients, are obtained based on impedance parameter simulation. =1N·s 2 / m, candidate damping coefficient =20 N·s / m, candidate stiffness coefficient =1000 N / m.

[0129] Step 5: Compare the actual fingertip gripping force with the gripping force threshold to determine the contact state between the dexterous hand and the fruits and vegetables. If the actual gripping force is less than the gripping force threshold, it means that the dexterous hand is in free space, and the output gripping force error is set. The value is 0; otherwise, the gripping force error is obtained by comparing the difference between the expected fingertip gripping force and the actual fingertip gripping force using a comparator. And obtain the rate of change of gripping force error ;

[0130] Fuzzy controller based on grasping force error and error change rate Change in output inertia coefficient Change in damping coefficient The impedance control module adjusts the actual gripping force based on the output of the fuzzy controller and the desired pose, velocity, and acceleration of the fingertip in Cartesian space coordinates, the actual pose, velocity, and acceleration of the fingertip in Cartesian space coordinates, and the set stiffness coefficient K and desired gripping force. The fuzzy variable impedance controller system block diagram is shown below. Figure 5 .

[0131] The above fuzzy controller is based on the grasping force error and error change rate Change in output inertia coefficient Change in damping coefficient include:

[0132] 51) Set the gripping force error range as follows: The rate of change of the gripping force error is within the range of The range of change in the coefficient of inertia is The range of damping coefficient variation is .

[0133] Input observations ( and ) and output control quantity ( and Each set is divided into seven fuzzy subsets: {NB, NM, NS, ZO, PS, PM, PB}. (Grasping force error) (Error rate of change) (Change in inertia coefficient) and The (dampening coefficient change) is fuzzified into a fuzzy subset: {negative large, negative medium, negative, zero, positive small, positive medium, positive large}, with corresponding records of {NB, NM, NS, ZO, PS, PM, PB}.

[0134] 52) Input fuzzification: Fuzzifying the input data using the grasping force error. and the rate of change of gripping force error (or ) Fuzzification to obtain the membership degree on different fuzzy subsets;

[0135] calculate Membership degree on 7 fuzzy subsets

[0136] (27)

[0137] calculate Membership degree on 7 fuzzy subsets

[0138] (28)

[0139] 53) The membership degree of the input observation is used for fuzzy inference according to the set fuzzy rules to obtain the fuzzy subsets corresponding to the changes in inertia coefficient and damping coefficient;

[0140] In this embodiment, based on the simulation results of the impedance control parameters obtained in step 4, fuzzy control rules for the impedance control parameters based on force control are formulated as shown in Tables 3 and 4.

[0141] (1) When and When it is large, Smaller size to improve response speed.

[0142] (2) When Larger but When smaller, Larger size to enhance stability.

[0143] (3) When and When approaching zero, .

[0144] (4) When or When it is large, Larger, to suppress oscillations.

[0145] (5) When and When smaller, Smaller size to improve response speed.

[0146] Table 3. Fuzzy Rule Table for Changes in Inertia Coefficient

[0147]

[0148] Table 4. Fuzzy Rule Table for Damping Coefficient Change

[0149]

[0150] Rule format:

[0151] (29)

[0152] That is: when the grasping force error e f Belongs to fuzzy sets And the rate of change of error Belongs to fuzzy set B h When the change in the output inertia coefficient ΔM belongs to the fuzzy set... Meanwhile, the change in damping coefficient ΔB belongs to a fuzzy set. ;

[0153] 54) Based on the candidate inertia coefficient and candidate damping coefficient Calculate the coefficient of inertia and damping coefficient The stiffness coefficient K will be set as the candidate stiffness coefficient. ;

[0154] The solutions for M and B are:

[0155] (30)

[0156] (31)

[0157] Set the stiffness coefficient K = 1000 N / m.

[0158] 55) Stiffness coefficient of fruits and vegetables Expected grasping power , Given that the desired pose can be obtained according to equation (26), After obtaining the desired pose value, the actual pose for controlling the dexterity hand movement adjustment can be solved by inverse kinematics of the dexterity hand. The impedance controller takes the inertia coefficient and damping coefficient output by the fuzzy controller, as well as the desired position, velocity, and acceleration in Cartesian space, and the fingertip position, velocity, and acceleration in Cartesian space, as inputs. Using dexterous hand kinematics, it adjusts the actual grasping force of the dexterous hand to approximate the desired grasping force. .

[0159] By traversing all combinations of input variables, inference and defuzzification are performed based on the fuzzy rule base, and the mapping relationship between input and output is plotted as a three-dimensional response surface, resulting in the three-dimensional response diagram of the fuzzy controller, as shown below. Figure 13 .

[0160] Triangle membership degree and linguistic variables, such as Figure 6 .

[0161] It should be noted that the specific embodiments described in this invention are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains can make various modifications or additions to the described specific embodiments or use similar methods to replace them, but without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A dexterous hand fruit and vegetable grasping control method based on fuzzy variable impedance control, characterized in that, Includes the following steps: Step 1: Construct a kinematic model of dexterous hand grasping fruits and vegetables, establish a multi-finger cooperative kinematic model based on spinor theory, and determine the pose of each fingertip in the base coordinate system, the grasping force of each finger, and the distribution of the resultant force of multiple fingers; Step 2: Based on force closure, construct a multi-objective gripping force optimization model with minimum contact force and minimum deviation between contact force and normal direction as the criterion. Solve the model using the NSGA-II algorithm to obtain the minimum gripping force required for stable and non-destructive gripping. Step 3: Establish an impedance control model in Cartesian space, and then simulate the established impedance control model to obtain candidate values ​​for impedance control parameters, including candidate inertia coefficients. Candidate damping coefficient Candidate stiffness coefficients ; Step 4: Introduce fuzzy control, using the grasping force tracking error and its rate of change as input, and outputting the inertia adjustment ΔM and damping adjustment ΔB, to construct a fuzzy adaptive variable impedance controller based on candidate inertia coefficients. and candidate damping coefficient Calculate the inertia coefficient of the fuzzy adaptive variable impedance controller and damping coefficient The stiffness coefficient K of the fuzzy adaptive variable impedance controller will be set as the candidate stiffness coefficient. Fuzzy adaptive variable impedance controller based on inertia coefficient and damping coefficient And using the desired pose, velocity, and acceleration of the fingertip in Cartesian space coordinates, the actual pose, velocity, and acceleration of the fingertip in Cartesian space coordinates, and the set stiffness coefficient K and desired gripping force as inputs, the system can achieve precise tracking of the gripping force. In step 1, a single-finger kinematic model is established using spinor theory, including: establishing the transformation relationship between the base coordinate system of each finger and the global coordinate system; using the exponential product formula to represent the fingertip pose; calculating the spatial Jacobian matrix of each finger; and establishing the mapping relationship between joint torque and contact force. In step 4, fuzzy control is introduced, taking the grasping force tracking error and its rate of change as input, and outputting the inertial adjustment amount ΔM and the damping adjustment amount ΔB, including: Grasping force error and error change rate The linguistic variables are fuzzed using the triangular membership function: {NB, NM, NS, ZO, PS, PM, PB}. Based on a preset fuzzy rule table, fuzzy inference is performed to obtain fuzzy outputs of the inertia change ΔM and the damping change ΔB; the fuzzy rule is: when the gripping force error e f Belongs to fuzzy sets And the rate of change of error Belongs to fuzzy set B h When the change in the output inertia coefficient ΔM belongs to the fuzzy set... Meanwhile, the change in damping coefficient ΔB belongs to a fuzzy set. ; Fuzzy set in the fuzzy rule table of inertia coefficient change The row corresponding to the middle subset and the fuzzy set B h The membership degree of a subset whose columns are orthogonal is a fuzzy set. A subset of; Fuzzy set in the fuzzy rule table of damping coefficient change The row corresponding to the middle subset and the fuzzy set B h The membership degree of a subset whose columns are orthogonal is a fuzzy set. A subset of; The fuzzy rule table for the change in inertia coefficient is as follows: The fuzzy rule table for the change in damping coefficient is as follows: 。 2. The method according to claim 1, characterized in that, In step 2, the gripping force optimization model includes the following constraints: force balance constraint, friction cone constraint, and joint torque constraint.

3. The method according to claim 1, characterized in that, In step 3, the impedance control model is as follows: In the formula, , , These are the expected inertia coefficient matrix, expected damping coefficient matrix, and expected stiffness coefficient matrix of the impedance model, respectively. This represents the actual pose of the robotic arm's end effector. The desired pose of the robotic arm's end effector. For actual grasping power, For the desired grasping force.

4. The method according to claim 3, characterized in that, The candidate inertia coefficient and candidate damping coefficient The desired stiffness coefficient matrix K represents the key adjustment parameters determined through simulation analysis. , .

5. A dexterous hand control system for implementing the method according to any one of claims 1-4, characterized in that, include: The kinematic modeling module is used to construct dexterous hand grasping models based on spinor theory. An optimized calculation module is used to solve for the minimum gripping force using the NSGA-II algorithm; Impedance control module is used to achieve initial tracking of the gripping force; The fuzzy inference module is used to dynamically adjust the impedance parameters M and B based on the grasping force error and the rate of change.

6. A fruit and vegetable harvesting robot, characterized in that, It includes the dexterous hand control system as described in claim 5.

Images