Joint torque estimation method and compliance control method of rope-driven five-fingered dexterous hand

Abstract

The invention provides a joint torque estimation method and a compliant control method for a rope-driven five-fingered dexterous hand. The estimation method comprises the following steps: constructing an incidence relation model among rope-driven muscle fiber force, motor control signals and motion parameters; and a multivariable Gaussian process regression algorithm is adopted to learn a mapping relation between the input characteristics and the joint torque based on the incidence relation model, joint torque estimation without an external force sensor is realized, and on this basis, compliance control of the five-fingered dexterous hand is realized through a force-position hybrid control method. According to the invention, by using related information of the motor body, force sensing and compliant control without additional sensors are achieved.

Description

Technical Field

[0001] This application relates to the field of motor control technology for robots, specifically to a method for estimating joint torque and a compliant control method for a rope-driven five-fingered dexterous hand. Background Technology

[0002] Dexterity is a key challenge in robotics. The human hand, with its high dexterity and precision, excels in tasks such as grasping and tool use, capabilities crucial for industrial production and daily life. Therefore, developing human-like dexterity robotic hands to achieve efficient grasping and tool manipulation has become a research hotspot at the intersection of robotics, intelligent manufacturing, and artificial intelligence. Among these technologies, motor-cable drive systems (mimicking human tendon structures) are an important solution for achieving high degrees of freedom in dexterity hands, but current technology still has significant shortcomings.

[0003] Rope-driven dexterous hands mimic tendon force through motors and cables, achieving high dexterity. Typical examples include the Shadow-hand and NASARobonant-2-hand, which can achieve 22 degrees of freedom (DOF), 20 of which are independently controlled. Despite their outstanding dexterity, these dexterous hands have significant deficiencies in the two core dimensions of sensory ability and compliant control, specifically including: Lack of sensory information: The complex structure makes it difficult to install force / tactile sensors on the joints / fingertips, resulting in a lack of accurate force feedback and affecting the precision of grasping control. Severe nonlinear interference: Nonlinear factors such as elastic deformation, friction, hysteresis, and cross-coupling of cables significantly reduce force transmission efficiency and control accuracy; Limitations of force control algorithms: Traditional algorithms cannot cope with system nonlinearity, resulting in poor force control accuracy during precise operation; Balancing flexibility and stiffness is difficult: When faced with objects of different hardness and shape, it is impossible to adaptively adjust control parameters, making it difficult to balance "gentle gripping" and "stable holding". Lack of overload protection: There is no effective joint overload protection mechanism, and inaccurate force control can easily lead to system damage.

[0004] To overcome the two obvious shortcomings mentioned above, there are two key technological challenges in rope-driven dexterity hands: 1. Challenges in sensory technology: Difficulty in integrating tactile perception Although the force conversion mechanism is well understood, embedding sensors, circuits, and communication devices into a compact rope drive system still presents significant challenges. Existing force sensing methods (relying on resistance, capacitance, and visual signals) have three major limitations: Sensor miniaturization is difficult: it is difficult to manufacture "small and accurate" sensors to be installed in joints / fingertips, which can easily affect the structure and dexterity of the hand; Wiring complexity: multi-joint structure leads to difficult cross-joint wiring, increasing system complexity and failure rate; Signal processing difficulty: system nonlinearity generates a large amount of noise and interference, and algorithm development is difficult.

[0005] 2. Soft control challenge: accurate and adaptive adjustment is difficult Soft control (let the hand "soft and stable" interaction) is a prominent shortcoming of the rope-driven system, and the specific problems include: Low force transmission efficiency: friction and elastic deformation cause large deviation between "input force" and "output force", making accurate force control difficult; Dynamic response is slow: slow response to external environmental changes and poor adaptability, making it difficult to achieve truly soft interaction; Overload protection is missing: traditional control methods have no effective joint overload protection, which can easily damage the system; Algorithm is not suitable: traditional PID control cannot cope with the complex dynamic characteristics of the system, making high-precision soft control difficult.

[0006] In summary, the core pain points of the rope-driven five-fingered dexterous hand are "weak sensing ability" and "poor soft control", and there is a lot of room for improvement in existing technology. SUMMARY

[0007] In view of the defects in the prior art, the purpose of the present application is to provide a joint torque estimation method and a soft control method for a rope-driven five-fingered dexterous hand.

[0008] The first aspect of the present application provides a joint torque estimation method for a rope-driven five-fingered dexterous hand, comprising: Constructing a correlation model between the equivalent output tension of the rope drive and the motor control signal and the motion parameters; Using a multivariate Gaussian process regression algorithm to learn the mapping relationship between the input features based on the correlation model and the joint torque, and to realize joint torque estimation without external force sensors.

[0009] Optionally, the correlation model is specifically: (1) In the formula, is the tendon force, i.e. the equivalent output tension; and represent the actual rope length and the desired rope length, respectively, and represent the actual rope speed and the desired rope speed, respectively, and represent the actual length and the initial preloaded length of the passive spring, both of which are motion parameters; represents a positive definite coefficient; I represents motor current, and is a motor control signal.

[0010] Optionally, the multi-variable Gaussian process regression algorithm is used to learn a mapping relationship between input features and joint torque based on the correlation model, so as to realize joint torque estimation without an external force sensor, including: Based on the correlation model, an equivalent output tension of the driving rope is calculated as an intermediate variable according to the motor control signal and the motion parameters at the current moment; Motor current, motor position, motor speed, joint position and joint speed are obtained, and an enhanced input feature vector is constructed in combination with the intermediate variable; Real joint torque labels corresponding to the enhanced input feature vector are obtained to form a sample data set, which is divided into a training set and a test set; A Gaussian process regression model is constructed, and a kernel function thereof is defined as a composite form; It is assumed that the output value of the Gaussian process regression model under a given training input obeys a Gaussian prior distribution with a mean of zero and a covariance matrix determined by the kernel function; By maximizing the marginal likelihood function, the hyperparameters in the kernel function are optimized, and the training of the Gaussian process regression model is completed; Based on the trained Gaussian process regression model, the actual joint torque is estimated with the enhanced input feature vector as input.

[0011] Optionally, the kernel function includes a radial basis function kernel and a white noise kernel.

[0012] Optionally, the real joint torque labels corresponding to the enhanced input feature vector are obtained, including: An external load test device is constructed, including an angle sensor, a fixed pulley, a cable and an adjustable mass block; the angle sensor is installed at the finger joint to measure angle information, the end of the finger to be tested is connected to the fixed pulley through the cable, and the mass block is hung on the finger joint through the pulley and the cable, for applying an external force to the specified joint; A mechanical transmission geometric relationship is constructed according to the initial state and the uniform motion state of the mass block, and a planar triangle model is established: triangle ABC is composed of the thumb joint A, the pulley anchor point B and the action point C of the cable at the end of the thumb; triangle BCE is composed of the pulley anchor point B, the action point C of the cable at the end of the thumb and the tangent point E of the cable and the pulley; triangle ACD is composed of the thumb joint A, the action point C of the cable at the end of the thumb and the intersection point D of the finger generatrix and the cable; Based on the geometric relationship of the triangle, the real joint torque is obtained in combination with the cosine theorem and force transmission analysis.

[0013] Optionally, based on the geometric relationship of the triangle, combined with the cosine theorem and force transmission analysis, the real joint torque is obtained, comprising: In triangle ABC, ; ; ; represents the baseline motion angle of the finger joint, represents the angle between the finger baseline and the joint-rope anchor point, represents an auxiliary angle in triangle ABC; represents the distance from the center of the pulley to the center of the finger joint, represents the distance from the joint origin to the foot of the rope on the finger baseline; In triangle BCE, the following relationship is obtained according to the cosine theorem: ; In triangle ACD, the angle between the cable and the finger generatrix is calculated as: ; The force arm is calculated as: ; When the weight moves at a constant speed under the traction of the cable, the external torque on the joint is:

[0014] The second aspect of the application provides a compliant control method for a rope-driven five-fingered dexterous hand, comprising: The joint torque obtained by any one of the joint torque estimation methods for the rope-driven five-fingered dexterous hand is used to construct an impedance mathematical model of mass, damping and spring; Based on the impedance mathematical model, the desired inertia matrix, damping matrix, stiffness matrix and external force information are substituted into the impedance control calculation to solve the joint angular acceleration first, and then the joint angular velocity and joint angle at the next time are obtained through iterative calculation; Based on the updated joint angle and angular velocity, the motor is driven to execute and perform compliant control.

[0015] Optionally, the impedance mathematical model is specifically: ; In the formula, are the desired joint torque and the load joint torque, respectively; is the desired inertia matrix; is the desired damping matrix; is a desired stiffness matrix; are a desired joint angle, angular velocity and angular acceleration, respectively; are an adjusted joint angle, angular velocity and angular acceleration under the action of an external force, respectively.

[0016] In a third aspect, the present application provides a terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor is configured to execute the method.

[0017] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the program is executable by a processor to execute the method.

[0018] The joint torque estimation method based on body perception of the rope-driven five-fingered dexterous hand provided by the present application does not need to rely on an external force sensor, extracts key features from motor body information by constructing a correlation model between muscle fiber force and motor control signals and motion parameters, and combines a multivariate Gaussian process regression algorithm to realize high-precision estimation of joint torque. This method effectively solves the problems of high cost, complex structure and poor reliability caused by the dependence of traditional dexterous hands on external sensors, and provides a feasible technical path for realizing low-cost and high-robustness robot body perception capability. The method of the present application is suitable for rope-driven five-fingered dexterous hands, and is particularly suitable for multi-joint robot hands with motor-cable coupling driving and bionic structure design.

[0019] Other technical effects brought by the additional features will be further described in the corresponding embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0020] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings: Figure 1 A flowchart of a joint torque estimation method for a rope-driven five-fingered dexterous hand according to an exemplary embodiment is shown. Figure 2 A geometric relationship diagram of data acquisition according to an exemplary embodiment is shown. Figure 3 A flowchart of a compliant control method according to an exemplary embodiment is shown. Figure 4 An impedance compliant control method block diagram of a joint torque estimation method according to an exemplary embodiment is shown. Figure 5 A comparison chart of the effect of hand force on different fingers (index finger IF, thumb TF) of a rope-driven dexterous hand under different prediction methods according to an exemplary embodiment is shown. Figure 6A comparison chart of the thumb (TF) and index finger (IF) pinch force-time response of a rope-driven dexterous hand according to an exemplary embodiment. DETAILED DESCRIPTION

[0021] The present application will be described in detail below with specific embodiments. The following examples will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. The parts not described in detail in the following examples can be implemented using existing technology.

[0022] In the prior art, the rope-driven five-fingered dexterous hand has obvious defects in sensing ability. Based on the above problems, the embodiment of the present application provides a joint torque estimation method for a rope-driven five-fingered dexterous hand based on proprioception to solve the above problems.

[0023] Reference Figure 1 As shown in the figure, in an embodiment of the present application, a joint torque estimation method for a rope-driven five-fingered dexterous hand includes: S100, constructing a correlation relationship model between the muscle fiber force of the rope drive and the motor control signal and the motion parameters; S200, using a multivariate Gaussian process regression algorithm to learn the mapping relationship between the input features based on the correlation relationship model and the joint torque, and realizing the joint torque estimation without external force sensor.

[0024] The above embodiment of the present application does not need to rely on external force sensors, but by constructing a correlation relationship model between muscle fiber force and motor control signal and motion parameters, and combining a multivariate Gaussian process regression algorithm, key features are extracted from motor body information to realize high-precision estimation of joint torque. This method effectively solves the problems of high cost, complex structure and poor reliability caused by relying on external sensors in traditional dexterous hands, and provides a feasible technical path for realizing low-cost and high-robustness robot proprioception ability.

[0025] The rope-driven structure imitates the muscle-tendon system of the human hand: the motor as the active driving unit simulates muscle contraction, the cable as the transmission medium simulates tendon force transmission, and the angle of the cable is designed to simulate the angle effect of the pinnate muscle fiber, optimize the force transmission efficiency, and realize natural and flexible movement similar to the human hand. Based on this structure, in order to avoid using external force sensors, in some specific embodiments of the present application, S100, a correlation relationship model between the muscle fiber force of the rope drive and the motor control signal and the motion parameters can be constructed using the following steps: S101: Based on the working mechanism of biological muscle structure, a Hill-type equivalent mechanical model is constructed to characterize the nonlinear dynamic characteristics of the rope-driven system.

[0026] Specifically, this step draws inspiration from the structure and functional mechanisms of the human muscle-tendon system, and biomimetically abstracts the motor-cable drive system. Specifically: the DC motor simulates muscle fibers, providing active contractile force; the cable simulates tendons, transmitting tension to the finger joints; and the elastic elements in the transmission path simulate series / parallel elastic tissues, absorbing deformation energy.

[0027] Based on this, a Hill-like numerical model is established, which includes three core components: Contraction element (CE): Driven by motor control signals, generating the main force; Series elastic element (SEE): Located in the cable path, it simulates the stretching and contraction of tendons and ligaments; Parallel elastic element (PEE): reflects lateral deformation or preload relaxation of the structure.

[0028] This model can describe the strong nonlinear coupling relationship between tendon force and cable length and speed under different loads, providing a physical basis for subsequent high-precision force estimation and control.

[0029] S102: Analyze the relationship between equivalent muscle force and motion state from the Hill-like model in S101, and identify and extract key parameters that affect system performance.

[0030] Specifically, during model operation, the magnitude of the equivalent "muscle fiber force" depends not only on the control input (such as current) but is also significantly influenced by the current mechanical state. Through dynamic analysis of the model's internal variables, the following parameters with key impacts on force output were identified: The deviation between the actual cable length and the expected length characterizes the degree of deformation of the series elastic element (SEE) and reflects the degree of elastic deformation. The difference between the actual cable speed and the expected speed reflects the dynamic inconsistency between the contraction element (CE) drive command and the actual response of the SEE, reflecting the hysteresis of friction and inertial response. The difference between the actual length of the passive spring segment and the initial preload length directly corresponds to the relaxation state of the parallel elastic element (PEE), reflecting the tension decay after long-term use of the system.

[0031] It is worth noting that the passive spring here does not refer to a specific mechanical component, but rather to the modeling abstraction of the equivalent elastic behavior in the rope-driven structure caused by factors such as structural flexibility, assembly preload, and material creep. Its length change reflects the tension degradation phenomenon that occurs in the system during long-term operation and can be used for the design of online compensation control strategies.

[0032] S103, based on the extracted key parameters, designs a targeted compensation mechanism for the elastic deformation, friction loss and hysteresis effect in the rope drive system.

[0033] Specifically, using the parameters identified in S102, an equivalent compensation model (relationship model) of the following form is constructed: ; In the formula, Tendon strength (muscle fiber strength); and These represent the actual rope length and the expected rope length, respectively. and These represent the actual rope speed and the desired rope speed, respectively. and These represent the actual length and the initial preload length of the passive spring, respectively. This represents the positive definite coefficient, and I is the motor current.

[0034] The embodiments described above in this application effectively suppress the effects of elastic deformation, friction, and hysteresis through biomimetic modeling and key parameter compensation, thereby improving the accuracy of cable tension estimation and enhancing the dynamic response stability and long-term operational reliability of the system.

[0035] Gaussian process regression (GPR) is widely used in data modeling and uncertainty estimation. It uses kernel functions to model complex, nonlinear relationships in multi-feature data, especially in high-dimensional spaces. GPR's uncertainty quantification improves reliability and supports data analysis, particularly in small-sample, high-dimensional scenarios, reducing overfitting and ensuring robust predictions. Based on this, to obtain more accurate torque estimation, in some specific embodiments of this application, S200, based on a correlation model, a multivariate Gaussian process regression algorithm is used to learn the mapping relationship between input features and joint torque, achieving joint torque estimation without external force sensors. This can be achieved through the following steps: S201, based on the correlation model obtained from S100, calculates the equivalent output tension in the cable according to the motor control signal and motion state at the current moment, as an intermediate variable; S202, obtain motor current, motor position, motor speed, joint position and joint speed, as well as intermediate variables, and construct an enhanced feature input vector; Specifically, the input vector can be standardized with zero mean and unit variance to improve the model's generalization ability.

[0036] S203, obtain a sample dataset containing the above-mentioned enhanced input features and corresponding real joint torque labels, and divide it into a training set X and a test set Y; ; in, This represents the enhanced feature input vector, corresponding to the motor current, motor position, motor speed, joint position, joint speed, and cable equivalent tension of the dexterous hand; Indicates and The corresponding actual joint torque; N is the total number of samples; The data is segmented as follows: ; This indicates that it is used to train the GPR model. This is used to verify the accuracy of the model; yes The corresponding actual torque, yes The corresponding actual torque; S204, performs zero-mean and unit-variance standardization on X and Y respectively: ; In the formula, Indicates the mean and sum of the data Indicates the standard deviation of the data. This represents the standardized input vector. This represents the standardized true torque label vector; S205 effectively handles the nonlinear characteristics and noise in rope drive systems by combining the RBF kernel function with the white noise kernel function.

[0037] Specifically, in GPR modeling, the kernel function is defined as: ; In the formula, It is the RBF kernel function, which represents the similarity between data points, and , is the white noise kernel function, representing the measurement noise. Indicates the current sample, Indicates historical samples; Indicates the magnitude weight of the RBF kernel. Indicates the length scale parameter; Represents the Dirac function, Represents the variance of white noise; S206, Assume the prior distribution is: ; Where K is the kernel matrix, and its elements are given by the kernel function k. . S207, given training data The predicted distribution is: ; The mean is The variance is I represents the identity matrix. Indicates the test point; Indicates the prediction result; and K The covariance matrices are calculated using the kernel function k(x,x'), and they represent the similarity relationships between the training and test data, respectively. The kernel function value of the test point itself; S208 optimizes the model by maximizing the marginal likelihood function to determine the hyperparameters. :

[0038] S209, Model predictions are made using posterior means. and standard deviation :

[0039] S210, the regression results are evaluated using metrics such as mean squared error (MSE) and coefficient of determination. : ;

[0040] The embodiments of this application construct enhanced features by fusing the equivalent tension output by the physical model with the motor / joint state, and combine standardized preprocessing and composite kernel function GPR modeling to effectively improve the accuracy of joint torque estimation.

[0041] like Figure 2 As shown, this paper presents a mechanical / biomimetic system (possibly a human-like finger or robotic arm) and its mechanical testing and structural analysis scheme, which consists of two parts: a test scenario and a mechanical model. Left side: Test scenario The joints involved in this test include: IF-PIP (proximal interphalangeal joint of the index finger), IF-MPR (abduction / rotation joint of the index finger at the metacarpophalangeal base), TF-MPP (flexion-extension joint of the thumb), TF-MPR (abduction / rotation joint of the thumb), TF-CMR (compound motor joint of the thumb), and TF-CMP (carpal joint of the thumb). Since all four fingers use the same modular design and have identical structure and function, only the index finger needs to be tested as a representative. The above tests target key performance indicators of the mechanical joints, such as torque, range of motion, and load-bearing capacity, and are verified by referring to the functions of the metacarpophalangeal (MP) and proximal interphalangeal (PIP) joints of the human fingers.

[0042] Right side: Mechanical model and structural annotations This section covers the force analysis and key component / parameter definition of the system, with the core being the simulation of the mechanical transmission of a "weight driving a mechanical structure through a rope-pulley system": The main components are as follows: Rope (C) and pulley (E): used to transmit tension; Weight (m) cp ): Provides driving force (gravity); Robotic arm or finger structure (including nodes A, B, etc.): the driven actuator.

[0043] Key parameters: Length parameter ( (etc.): Dimensions / lever arm length of each component; Angular parameters ( etc.): Angle between components (decomposition of influence); force( ): The tension transmitted by the rope.

[0044] based on Figure 3 In some specific embodiments of this application, a method for obtaining the corresponding actual joint torque is provided. The specific steps are as follows: S2031: Set up an external load testing device.

[0045] Construct a testing system comprising a high-precision angle sensor, fixed pulleys, traction cables, and an adjustable mass. The tip of the finger to be tested is connected to the pulley system via a tendon cord, and the mass is suspended from the extension of the tendon cord via the pulleys, thereby applying a controllable external tension to the specified joint.

[0046] in: High-precision angle sensors are used to measure joint rotation angles in real time, providing a basis for geometric modeling and motion state judgment; Fixed pulleys are used to change the direction of force transmission, ensuring a stable and controllable loading path; The traction cable serves as the force transmission medium, reliably transmitting the tensile force generated by the mass block. Adjustable mass blocks provide variable yet constant gravity loads to simulate stress conditions under different contact conditions.

[0047] S2032: Collect joint status and determine steady state.

[0048] A high-precision angle sensor installed at the joint is used to collect the current angle position of the finger in real time, and the motion feedback is combined to determine whether the system has entered a steady state or a uniform motion state, so as to ensure minimal dynamic interference.

[0049] S2033: Entering a quasi-static equilibrium state.

[0050] When the mass block moves slowly at a constant speed under the traction of the tendon rope, the effects of inertia and acceleration are ignored, and the entire transmission system is considered to be in a quasi-static equilibrium state. At this time, the external force is determined only by gravity and geometry.

[0051] S2034: Establish a plane geometric triangle model.

[0052] Based on the mechanical transmission path, three key planar triangles are constructed for analyzing the force transmission path: △ABC: It is composed of the thumb rotation joint point A, the pulley anchor point B, and the point of action of the tendon rope at the end of the thumb C (anchor point); △BCE: Consists of pulley anchor point B, point of action C, and the point of tangency between the tendon rope and the pulley E; △ACD: Composed of joint point A, point of action C, and intersection point D of the finger generatrix and tendon ligament projection.

[0053] S2035: Calculates the actual joint torque based on geometric relationships.

[0054] Specifically, such as Figure 2 As shown, in △ABC, ; ; ; Indicates the baseline movement angle of the finger joint. This indicates the angle between the finger baseline and the line connecting the joint and the rope anchor point. Indicates an auxiliary angle within triangle ABC; This indicates the distance from the center of the pulley to the center of the finger joint. It represents the distance from the origin of the joint to the foot of the rope perpendicular to the finger baseline; In triangle BCE, the following relationship is obtained according to the Law of Cosines: ; △ACD, the angle between the cable and the finger generatrix is ​​calculated as follows: ; lever arm The calculation is as follows: ; When the weight moves at a constant speed under the traction of the cable, the external torque on the joint is:

[0055] Through the above embodiments, by combining geometric modeling and quasi-static calibration with high-precision angle feedback, highly repeatable measurement of real joint torque is achieved, providing reliable training data.

[0056] Based on the same technical concept, other embodiments of this application provide a compliant control method for a rope-driven five-finger dexterity hand, such as... Figure 3 As shown, it includes the following steps: S10 uses the joint torque estimation method of rope-driven five-finger dexterity hand to obtain the joint torque, and constructs the impedance mathematical model of mass, damping and spring. S20, based on the impedance mathematical model, substitutes the desired inertia matrix, damping matrix, stiffness matrix and external force information to perform impedance control calculation. First, the joint angular acceleration is obtained, and then the joint angular velocity and joint angle at the next moment are obtained through iterative calculation. S30 inputs the updated joint angles and angular velocities as compliant trajectory inputs to the underlying position / velocity controller, driving the motor to perform the corresponding adaptive motion.

[0057] In the embodiments described above, the precise force control strategy without additional force sensors achieves high-precision force feedback estimation by utilizing proprioceptive information to replace direct force sensing. Simultaneously, the established mapping relationship between fingertip force and joint torque further ensures force control accuracy, thus enabling precise operation in contact detection and force magnitude control without the need for additional sensors. Furthermore, the adaptive overload protection mechanism automatically adjusts control parameters based on real-time torque estimation to prevent joint overload, achieves smooth transitions during changes in contact state to avoid control system oscillations, and automatically adjusts control stiffness for objects of varying hardness, thereby balancing compliance and stability.

[0058] In some specific implementations, the impedance mathematical model is as follows: ; In the formula, These are the desired joint torque and the loaded joint torque, respectively. It is the expected inertia matrix; It is the desired damping matrix; It is the desired stiffness matrix; These are the desired joint angle, angular velocity, and angular acceleration, respectively. These are the joint angle, angular velocity, and angular acceleration adjusted under the action of external forces.

[0059] In some specific implementations, S20 incorporates the estimated joint torque information into the control system through a discretization control algorithm, specifically as follows: ;

[0060] The embodiments described above in this application dynamically solve joint acceleration and iteratively update motion state through an impedance model, thereby achieving an adaptive response to environmental interaction forces and improving dexterity and operational safety.

[0061] Based on the joint torque estimation method and compliant control method described above, some specific embodiments of this application provide an advanced robot joint control process that combines Gaussian process regression (GPR), admittance control, and underlying PID motor control. As shown in Figure 4, this control process architecture is suitable for the "Dex-Hand021" dexterous hand shown in the figure, and is used to handle complex force interaction tasks.

[0062] S1: Data Acquisition from the Perception Layer S1.1 collects key state parameters of the physical joint in real time, including joint angle, motor angle, and motor current.

[0063] S2: Nonlinear Modeling and Moment Estimation using the GPR Module S2.1 Input data: Input the motor current, angle and other state variables collected in step 1 into the Gaussian process regression (GPR) model.

[0064] S2.2 Calculate the predicted mean using the formula Calculate the predicted mean (i.e., Y) pred ).

[0065] S2.3 Calculate the prediction variance using the formula .

[0066] S3: Admittance control module compliance adjustment S3.1 The torque deviation is obtained based on the predicted value output by the GPR module. Through formula This is converted into the corresponding torque τ.

[0067] S3.2 converts the torque and the Jacobian matrix J based on the motor angle. ， Input the transfer function of the second-order mass-damped-spring system , Indicates virtual quality. Indicates virtual damping. Indicates virtual stiffness, It is the Laplace operator.

[0068] S3.3 obtains the angle correction amount Δθ through transfer function calculation, so that the robot arm actively generates displacement offset when subjected to force, exhibiting spring-like compliance.

[0069] S4: Closed-loop control of actuator motors and joints S4.1 Joint Control (Force-Position Hybrid Control): S4.1.1 Start PID control: Enable standard position loop Integral ring and differential ring .

[0070] S4.1.2 Synthesize the final instruction: Combine the initial target instruction The final joint reference position is obtained by adding the Δθ generated by admittance control.

[0071] S4.2 Motor Control (Underlying Current Loop Control): S4.2.1 uses a tendon-driven motor-rope model to convert the rotational motion of the motor into the tension of the rope.

[0072] S4.2.2 Based on the control law Calculate the motor output current.

[0073] Indicates the expected length of the tendon ligament. Indicates the actual length of the tendon ligament. This represents the expected rate of change in the length of the tendon ligament. This represents the actual rate of change of length. Indicates the reference length of the spring. This indicates the measured length of the tendon cord on the motor side. The proportional gain represents the length deviation. The differential gain represents the rate of change of length. This indicates the compensatory gain of the chordae tendon elasticity.

[0074] S4.2.3 Combines the position feedback from the motor encoder to ensure accurate current output to drive the motor and complete position closed-loop control.

[0075] The preferred features in the above embodiments can be used individually in any embodiment, or in any combination thereof, provided they do not conflict with each other. Furthermore, parts not described in detail in the embodiments can be implemented using existing technologies.

[0076] The following examples and comparative examples will be used to further illustrate this application in order to better understand the above-mentioned technical solutions. It should be understood that the following are only some examples and are not intended to limit this application.

[0077] Experimental Example 1: This experimental case investigated torque estimation for six joints of a modular five-finger system, where the IF (index finger) configuration was applied to all four fingers. Ground truth data collection experiments were conducted for the six different joints.

[0078] For each joint, loads ranging from 0.05 kg to 1 kg were applied, with experiments conducted in 0.05 kg increments. Each load condition was tested for 30 seconds, with a sampling frequency of 100 Hz. This method generated a total of 360,000 sample groups.

[0079] The verification results show that, Figure 5 and Figure 6 As shown, the mean squared errors (MSEs) for IF-PIP (proximal interphalangeal joint of the index finger), IF-MPR (abduction / rotation joint of the metacarpophalangeal base of the index finger), TF-MPP (metacarpophalangeal flexion-extension joint of the thumb), TF-MPR (abduction / rotation joint of the thumb), TF-CMR (compound motor joint of the thumb), and TF-CMP (compound motor point or metacarpophalangeal joint of the thumb) are 0.0006 Nm, 0.0005 Nm, 0.0005 Nm, 0.0007 Nm, 0.0065 Nm, and 0.0066 Nm, respectively. The corresponding coefficients of determination (R²) are 0.9843, 0.9743, 0.9721, 0.9541, 0.9512, and 0.9533.

[0080] Experiment Example 2 This experimental example uses the Dex-Hand021 to evaluate the impedance control method compared to traditional PID control. The experiment involved grasping objects of varying hardness, size, and shape (wooden blocks, simulated fruit, water bottles, paper cups, tissue packs, and plush toys), with 30 trials per object to ensure statistical reliability. An impedance controller utilizing joint torque estimation automatically stopped the motor output upon contact to ensure safe grasping. Results show that while both methods achieve stable grasping, traditional PID control requires a significantly larger joint angle. Quantitative analysis reveals that, for objects ranging from hard to soft, the proposed method reduces joint energy consumption by 40.52%, 38.36%, 37.04%, 29.82%, 22.77%, and 18.6%, respectively, with an average reduction of 31.19%. This improved performance is achieved without an additional force sensor. The proposed method provides accurate force feedback estimation and dynamic adjustment, improving safety, stability, and control precision while reducing energy consumption and extending hardware lifespan. These advantages are particularly evident in high-frequency operation and interaction with rigid environments.

[0081] Based on the same technical concept, in other embodiments of this application, a terminal is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it can be used to perform the methods described above.

[0082] Based on the same technical concept, in other embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, can be used to perform the above-described method.

[0083] Optionally, the memory is used to store programs; the memory may include volatile memory, such as random-access memory (RAM), such as static random-access memory (SRAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc.; the memory may also include non-volatile memory, such as flash memory. The memory is used to store computer programs (such as application programs, functional modules, etc. that implement the above methods), computer instructions, etc., and the aforementioned computer programs, computer instructions, etc., can be partitioned and stored in one or more memories. Furthermore, the aforementioned computer programs, computer instructions, data, etc., can be accessed by the processor.

[0084] The aforementioned computer programs, computer instructions, etc., can be stored in partitions within one or more memory locations. Furthermore, the aforementioned computer programs, computer instructions, data, etc., can be accessed by a processor.

[0085] A processor is used to execute a computer program stored in memory to implement the various steps of the methods involved in the above embodiments. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0086] The processor and memory can be separate structures or integrated structures. When the processor and memory are separate structures, they can be coupled together via a bus.

[0087] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0090] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0091] The foregoing has described some specific embodiments of this application. It should be understood that this application is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the substantive content of this application. The above-described preferred features can be used in any combination without conflict.

Claims

1. A method for estimating joint torque in a rope-driven five-finger dexterity hand, characterized in that, include: Construct a model relating the equivalent output tension of the rope drive to the motor control signal and motion parameters; A multivariate Gaussian process regression algorithm is used to learn the mapping relationship between input features and joint torque based on the aforementioned correlation model, thereby achieving joint torque estimation without the need for external force sensors.

2. The method for estimating joint torque in a rope-driven five-fingered dexterous hand according to claim 1, characterized in that, The relationship model is specifically as follows: ； In the formula, It is tendon force, that is, equivalent output tensile force; and These represent the actual rope length and the expected rope length, respectively. and These represent the actual rope speed and the desired rope speed, respectively. and These represent the actual length and initial preload length of the passive spring, respectively, both of which are motion parameters; Indicates positive definite coefficients; I represents the motor current, which is the motor control signal.

3. The method for estimating joint torque in a rope-driven five-fingered dexterous hand according to claim 1, characterized in that, The method employs a multivariate Gaussian process regression algorithm to learn the mapping relationship between input features and joint torque based on the correlation model, thereby achieving joint torque estimation without the need for external force sensors. This includes: Based on the aforementioned correlation model, the equivalent output tension of the drive rope is calculated as an intermediate variable according to the motor control signal and motion parameters at the current moment. The motor current, motor position, motor speed, joint position, and joint speed are obtained, and combined with the intermediate variables, an enhanced input feature vector is constructed. Obtain the real joint torque labels corresponding to the enhanced input feature vectors, form a sample dataset, and divide it into a training set and a test set; Construct a Gaussian process regression model and define its kernel function as a composite form; Assume that, given the training input, the output of the Gaussian process regression model follows a Gaussian prior distribution with a mean of zero and a covariance matrix determined by the kernel function. The training of the Gaussian process regression model is completed by maximizing the marginal likelihood function and optimizing the hyperparameters in the kernel function. Based on the trained Gaussian process regression model, the actual joint torque is estimated using the enhanced input feature vector as input.

4. The method for estimating joint torque in a rope-driven five-fingered dexterous hand according to claim 3, characterized in that, The kernel functions include radial basis function kernels and white noise kernels.

5. The method for estimating joint torque in a rope-driven five-fingered dexterous hand according to claim 3, characterized in that, The step of obtaining the true joint torque label corresponding to the enhanced input feature vector includes: An external load testing device is constructed, including an angle sensor, a fixed pulley, a cable, and an adjustable mass block. The angle sensor is installed at the finger joint to measure angle information. The end of the finger to be tested is connected to the fixed pulley via a cable, and the mass block is suspended from the finger joint via the pulley and cable to apply external force to the specified joint. Based on the initial state and the uniform motion of the mass block, construct the geometric relationship of the mechanical transmission and establish a planar triangle model: triangle ABC is formed by the thumb joint point A, the pulley anchor point B, and the cable anchor point C at the end of the thumb; triangle BCE is formed by the pulley anchor point B, the cable anchor point C at the end of the thumb, and the tangent point E between the cable and the pulley; triangle ACD is formed by the thumb joint point A, the anchor point C at the end of the thumb, the intersection point D of the finger generatrix and the cable. Based on the geometric relationship of the triangle, and combined with the law of cosines and force transmission analysis, the actual joint torque is obtained.

6. The method for estimating joint torque in a rope-driven five-fingered dexterous hand according to claim 5, characterized in that, Based on the geometric relationship of the triangle, and combined with the law of cosines and force transmission analysis, the actual joint torque is obtained, including: In triangle ABC, ； ； ； Indicates the baseline movement angle of the finger joint. This indicates the angle between the finger baseline and the line connecting the joint and the rope anchor point. Indicates an auxiliary angle within triangle ABC; This indicates the distance from the center of the pulley to the center of the finger joint. It represents the distance from the origin of the joint to the foot of the rope perpendicular to the finger baseline; In triangle BCE, the following relationship is obtained according to the Law of Cosines: ； In triangle ACD, the angle between the cable and the finger generatrix is ​​calculated as follows: ； lever arm The calculation is as follows: ； When the weight moves at a constant speed under the traction of the cable, the external torque on the joint is: 。 7. A compliant control method for a rope-driven five-finger dexterity hand, characterized in that, include: Using the joint torque estimation method for the rope-driven five-finger dexterity hand as described in any one of claims 1-6, a mathematical model of mass, damping, and spring impedance is constructed. Based on the aforementioned impedance mathematical model, the desired inertia matrix, damping matrix, stiffness matrix, and external force information are substituted to perform impedance control calculations. First, the joint angular acceleration is obtained, and then the joint angular velocity and joint angle at the next moment are obtained through iterative calculations. Based on the updated joint angles and angular velocities, the drive motor is executed to achieve compliant control.

8. The method for compliant control of a rope-driven five-finger dexterity hand according to claim 7, characterized in that, The impedance mathematical model is as follows: ； In the formula, These are the desired joint torque and the loaded joint torque, respectively. It is the expected inertia matrix; It is the desired damping matrix; It is the desired stiffness matrix; These are the desired joint angle, angular velocity, and angular acceleration, respectively. These are the joint angle, angular velocity, and angular acceleration adjusted under the action of external forces.

9. A terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it can be used to perform the method of any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, this program can be used to perform the method of any one of claims 1-8.

Images