Hand exoskeleton control system and method for piano on-demand auxiliary teaching

Abstract

The invention discloses a hand exoskeleton control system and method for piano on-demand auxiliary teaching. The hand exoskeleton control system comprises a hand exoskeleton mechanical module, a dynamics module, an outer ring admittance controller, an inner ring position controller and an online compensation unit. The hand exoskeleton mechanical module is composed of a servo motor, a dynamic torque sensor, a hand exoskeleton base, a shaft sleeve, a palm fixing plate and a finger module. The dynamics module is used for constructing a dynamics simplified model of the finger module; the outer ring admittance controller is used for generating a compliant trajectory compensation amount conforming to the actual dynamic response; the inner ring position controller is used for dynamically adjusting the output torque of the servo motor, reducing buffeting of the sliding mode and realizing finger playing trajectory tracking; and the online compensation unit is used for performing real-time estimation and feed-forward compensation on a nonlinear disturbance term existing in the tracking error of the inner loop position controller based on the RBF neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of educational robots, and particularly to a control system for an extra - hand exoskeleton for on - demand piano assisted teaching. Background Art

[0002] Piano playing, as a professional hand motor skill, has strict requirements on the posture of the human hand during playing [1] , the speed of finger hitting the keys when playing the piano [2] and the strength [3] etc. Learners need to gradually master these skills through a large number of repetitive trainings under the guidance of a piano teacher. When playing the piano, it is necessary to maintain the correct hand posture to ensure the finger hitting strength and flexibility, so as to play beautiful tones. For these repetitive trainings, an extra - hand exoskeleton can be designed to replace the teacher, and different control schemes can be designed to achieve the teaching effect of the extra - hand exoskeleton.

[0003] Currently, extra - hand exoskeletons are mainly used for rehabilitation assistance training and daily movement assistance of the hand. For such extra - hand exoskeletons, control strategies are the key points to ensure the effectiveness, comfort, and safety of the exoskeleton system assistance. The mainstream exoskeleton control strategies include position control based on predefined trajectories, also known as trajectory tracking control. In reference trajectory tracking control, the pre - recorded joint trajectory patterns of a person are usually used as the reference trajectory of the exoskeleton control system. To improve the usability and flexibility of the controller, the desired joint trajectories are usually parameterized according to the physical parameters of different users, and the control goal is to minimize the position error between the actual movement trajectory of the human - exoskeleton system and the desired trajectory. [4] . When the joint moves, in addition to the movement trajectory, the joint torque curve is also similar. Therefore, some researchers have developed control strategies based on predefined torques for single - joint exoskeletons [5] . Control strategies based on predefined torques often need to pre - adjust the torque table or torque curve according to different wearers. Due to the limitations of the predefined torque table, some learning methods also need to be introduced to increase the self - adaptability of the system to the assistance of the un - predefined part. The above control methods can be collectively referred to as passive control.

[0004] For the piano assisted teaching scenario, if the user can participate more in the assisted movement, the teaching effectiveness will be improved. It is expected that during the user's playing, the exoskeleton robot will consider more the deviation of the user's playing trajectory and only make up for the user's wrong part, thus reducing the intervention in the user. Thus, the "Assist - as - needed (AAN) strategy" is introduced [6] . The goal of assist - as - needed is to enable the user to participate more in the movement to achieve a better teaching assistance effect.

[0005] In summary, the following problems exist in the existing piano assistance devices and methods:

[0006] Insufficiently precise auxiliary control: Most of the existing hand exoskeleton control methods adopt fixed control parameters and fail to dynamically adjust according to the user's real-time needs, which easily causes over-assistance or under-assistance and lacks the ability to control on demand.

[0007] Poor stability of trajectory control: Some current methods do not consider the influence of external disturbances and non-linear uncertainties, and the control algorithm lacks a robust compensation mechanism, resulting in unstable trajectory control and low accuracy during the assistance process.

[0008] Insufficient human-computer interaction: There is a lack of a fine recognition and response mechanism for human-computer interaction forces, which may cause mechanical forced driving, poor user experience, and is likely to lead to discomfort or interference with autonomous training.

[0009] Unable to achieve compliant control: The existing control methods cannot achieve the flexible response of the mechanical system to external contact forces, and still forcibly pull when the user has a subjective movement intention, suppressing the user's autonomous participation.

[0010] Lack of intelligent adaptive ability: Most control methods cannot optimize parameters online according to the dynamic behavior of the user's fingers, do not have the ability of self-learning or self-adaptation, and are difficult to adapt to individual differences or changes in training stages.

[0011] Complex mechanical structure or high cost: In order to achieve independent driving of multiple fingers, some exoskeleton systems adopt a multi-motor configuration, resulting in a complex system structure, high cost, large volume, and inconvenient wearing.

[0012] [1] Liu Pingping. Discussion on the Piano Teaching for Preschool Education Majors [D]. Shijiazhuang: Hebei Normal University, 2011.

[0013] [2] Zhang Yifan. Empirical Research on the Piano Skills of "Speed Stability" and "Chord Uniformity" [D]. Chongqing: Southwest University, 2018.

[0014] [3] Lei Tingting. Application of Touch Techniques in Piano Performance and Perfect Embodiment of Tone Effects [J]. Home Drama, 2021(26): 63-64.

[0015] [4]Shimada H,Hirata T,Kimura Y,et al.Effects of a robotic walkingexercise on walkingperformance in community-dwelling elderly adults[J].Geriatrics&gerontology intemnational.2009,9(4):372-381.

[0016] [5]Shepherd M K,Rouse E J.Design and validation of a torque-controllable knee exoskeleton for sit-to-stand assistance[J].lEEE / ASMETransactions on Mechatronics,2017,22(4):1695-1704.

[0017] [6]Luo L,Peng L,Wang C,et al.A greedy assist-as-needed controller forupperlimb rehabilitation[J].IEEE transactions on neuralnetworks andlearningsystems,2019,30(11):3433-3443 Summary of the Invention

[0018] The purpose of the present invention is to overcome the deficiencies in the prior art, solve various problems of traditional piano teaching exoskeleton assistance, and provide a hand exoskeleton control system for on-demand piano-assisted teaching. Based on the admittance control principle, by constructing a dynamic admittance controller for finger movement, it can real-time sense the dynamic touch of the performer on the keys, dynamically adjust the admittance parameters of the exoskeleton joints, realize the adaptive matching of the touch resistance and the assisting torque, improve the accuracy of fingering training while ensuring the naturalness of the performance, and reduce the interference of human-machine collaborative movement.

[0019] The purpose of the present invention is achieved through the following technical solutions:

[0020] A hand exoskeleton control system for on-demand piano-assisted teaching, comprising a hand exoskeleton mechanical module, a dynamics module, an outer-loop admittance controller, an inner-loop position controller, and an online compensation unit connected in sequence;

[0021] The hand exoskeleton mechanical module consists of a servo motor (1), a dynamic torque sensor (2), a hand exoskeleton base (3), a bushing (4), a palm fixing plate (5), and a finger module (6);

[0022] The servo motor (1) and the dynamic torque sensor (2) are fixed on the hand exoskeleton base (3), the palm fixing plate (5) is fixed on the upper surface of the hand exoskeleton base (3), the transmission shaft of the servo motor (1) is connected to the input shaft of the dynamic torque sensor (2) through a coupling, the output shaft of the dynamic torque sensor (2) is fixed to the bushing (4) through a flat key, a Bowden cable is fixed on the bushing (4), and the bushing (4) is connected to the finger module (6) through the Bowden cable to realize the transmission of the power of the servo motor to the finger module 6;

[0023] The dynamics module is used to construct a dynamic simplified model of the finger module (6);

[0024] The outer loop admittance controller is used to adaptively adjust the equivalent stiffness and damping parameters of the outer loop admittance controller based on the deviation between the actual human-machine interaction force and the ideal human-machine interaction force, combined with the inertia and damping characteristics in the dynamic simplified model, and generate a compliant trajectory compensation amount that conforms to the actual dynamic response;

[0025] The inner loop position controller is used to take the compliant trajectory compensation amount as the tracking target, construct a composite control law combining the feedforward compensation term and the sliding mode surface function through the inverse dynamics calculation of the dynamic simplified model, and dynamically adjust the output torque of the servo motor based on the exponential reaching law function to reduce the sliding mode chattering and realize the finger playing the piano trajectory tracking;

[0026] The online compensation unit is used to perform real-time estimation and feedforward compensation on the non-linear disturbance term existing in the tracking error of the inner loop position controller based on the RBF neural network.

[0027] Furthermore, the finger module (6) includes a middle phalanx fixing plate, a proximal phalanx top plate, a proximal phalanx bottom plate, bilateral proximal phalanx side links, and a drive shaft fixing plate,

[0028] One end of the drive shaft fixing plate is connected to the palm fixing plate, and the other end is hinged to the proximal phalanx bottom plate through a rotating shaft. The proximal phalanx bottom plate and the proximal phalanx top plate are connected to each other through bilateral proximal phalanx side links to form a parallelogram linkage mechanism. One end of the proximal phalanx top plate is hinged to the middle phalanx fixing plate through a rotating shaft, and a magic tape is provided on the middle phalanx fixing plate for transmitting the movement to the finger proximal phalanx and / or the middle phalanx;

[0029] Both ends of the bilateral proximal phalanx side links are respectively hinged to the proximal phalanx top plate and the proximal phalanx bottom plate to realize the parallel displacement of the proximal phalanx top plate or the phalanx bottom plate;

[0030] An angle sensor is also integrated on the near phalanx bottom plate to detect the joint angle in real time. A fixed retaining ring is also provided on the near phalanx top plate to limit the rotation angle range of the bilateral near phalanx side linkages to -30° to 40°, preventing joint overload.

[0031] The near phalanx bottom plate is connected to the bushing (4) through a Bowden cable. The torque output by the servo motor (1) is transmitted to the near phalanx bottom plate of the finger module (6) through the Bowden cable.

[0032] Furthermore, the outer loop admittance controller uses fuzzy rules to achieve adaptive adjustment of the admittance parameters. Taking the trajectory error and the human-machine interaction force as input variables, it outputs an online adjustment strategy for the admittance parameters. The fuzzy rules adopt the Mamdani algorithm, the membership function is a Gaussian function, and the admittance parameters include equivalent stiffness and damping.

[0033] Furthermore, the RBF neural network has several input layers, several hidden layers and an output layer, and updates the weights of the hidden layer online through an adaptive law to approximate the uncertain terms of the inner loop position control.

[0034] The present invention also provides a method for on-demand assisted control of a hand exoskeleton. Based on the hand exoskeleton control system described above, it includes:

[0035] Real-time collect the human-machine interaction force through a dynamic torque sensor, and perform a double-threshold judgment with a preset ideal interaction force;

[0036] When the interaction force error exceeds the threshold, trigger the outer loop admittance controller, dynamically adjust the stiffness and damping parameters, and generate a compliant trajectory compensation amount;

[0037] Input the compliant trajectory compensation amount into the inner loop position controller, and calculate and output the required driving torque according to the sliding mode surface function and the exponential reaching law function;

[0038] Use the RBF neural network to online estimate the non-linear disturbance term in the position tracking error, and use the estimated value as a feedforward term for compensation;

[0039] Continuously and circularly execute the above steps to realize on-demand assisted piano playing assistance.

[0040] Compared with the prior art, the beneficial effects brought by the technical solution of the present invention are:

[0041] 1. Precise on-demand assistance: The dynamic torque sensor real-time detects the human-machine interaction force, and forms a double-threshold judgment mechanism with the trajectory error judgment. Only when the user's playing deviation exceeds the threshold, the auxiliary torque is output, avoiding the over-reliance on the traditional full-course forced traction phenomenon. At the same time, it specifically strengthens the weak links of uneven strength and lagging rhythm, thereby improving the training efficiency and action standardization.

[0042] 2. High-robust trajectory tracking: The inner-loop position controller introduces a linear sliding mode surface and an exponential reaching law, enabling it to suppress uncertainties and external disturbances, ensuring fast convergence and high-precision tracking of joint trajectories even in complex dynamic environments.

[0043] 3. Online disturbance compensation: The RBF neural network uses Gaussian basis functions to approximate nonlinear uncertain terms and, combined with an adaptive weight update law, can compensate for unknown disturbances such as friction and modeling errors in real time, significantly improving control stability and comfort.

[0044] 4. Human-robot compliant interaction: The fuzzy adaptive outer-loop admittance controller adjusts the system stiffness and damping online according to the combined state of the trajectory error and interaction force, achieving a smooth transition from "soft following" to "rigid correction", dynamically matching the user's active movement intention, and taking into account both naturalness and accuracy, which helps to maintain a trainee-led movement experience.

[0045] 5. Lightweight and low-cost structure: Using only a single servo motor with Bowden cable drive can drive multiple finger modules simultaneously, simplifying the mechanical structure, reducing costs and weight, and improving wearing comfort and system maintainability.

[0046] 6. Multi-source data fusion: By combining the feedback of dynamic torque sensors and angle sensors, a finger dynamics admittance model and a trajectory generation model can be accurately constructed to achieve closed-loop fine adjustment of control, enhancing the learning immersion and physiological compatibility of the system.

[0047] In summary, the control system of the present invention is a high-precision human-computer interaction system. Using a single servo motor to drive the hand exoskeleton to drive the MCP joint and PIP joint of the human hand to perform piano-playing movements, thereby realizing the function of auxiliary training. Through multi-source data fusion of dynamic torque sensors and angle sensors, compliant trajectory compensation based on the outer-loop admittance controller, combined with robust trajectory tracking of the inner-loop position control and RBF neural network disturbance compensation, the synchronization of finger module movements with finger joint movements is achieved. In addition, it can also suppress the uncertainties and external disturbances of the inner-loop position controller, ensure a high degree of matching between the exoskeleton movements and natural human movements, and significantly improve the physiological compatibility and learning immersion of piano key-touching actions. Brief Description of the Drawings

[0048] Figure 1 is a schematic structural diagram of the hand exoskeleton mechanical module of the present invention;

[0049] Figure 2 is a schematic structural diagram of the driving part of the hand exoskeleton mechanical module of the present invention;

[0050] Figure 3 is a schematic diagram of the Bowden cable drive principle of the present invention;

[0051] Figure 4It is a schematic diagram of the finger module structure of the present invention;

[0052] Figure 5 It is a schematic diagram of the dynamic simplified model of the finger module of the present invention;

[0053] Figure 6 It is a schematic diagram of the process of the inner and outer loop control links in the hand exoskeleton control system of the present invention;

[0054] Figure 7 It is a schematic diagram of the RBF neural network structure of the present invention;

[0055] Figure 8 It is a joint trajectory tracking diagram realized by the control system of the present invention;

[0056] Figure 9 It is a comparison diagram of the joint trajectory tracking error realized by the control system of the present invention.

[0057] Figure 10 It is a compensation estimation diagram of the unknown terms of the control system by the RBF neural network of the present invention.

[0058] Figure 11 It is a diagram of the on-demand assistive force of the hand exoskeleton and the change of the actual moment of the human hand of the present invention. Detailed implementation manners

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0060] This embodiment provides a hand exoskeleton control system for on-demand assisted teaching of the piano, including a hand exoskeleton mechanical module, a dynamics module, an outer loop admittance controller, an inner loop position controller, and an online compensation unit;

[0061] (1) As shown in Figures 1 to 3 , the hand exoskeleton mechanical module is composed of a servo motor 1, a dynamic torque sensor 2, a hand exoskeleton base 3, a bushing 4, a palm fixing plate 5, and a finger module 6;

[0062] The servo motor 1 and the dynamic torque sensor 2 are fixed on the base 3 of the hand exoskeleton. The palm fixing plate 5 is fixed on the upper surface of the base 3 of the hand exoskeleton. The transmission shaft of the servo motor 1 is connected to the input shaft of the dynamic torque sensor 2 through a flexible coupling. The output shaft of the dynamic torque sensor 2 is fixed to the shaft sleeve 4 through a flat key. A Bowden wire is fixed on the shaft sleeve 4, and the shaft sleeve 4 is connected to the proximal phalanx bottom plate of the finger module 6 through the Bowden wire, realizing the transmission of the power of the servo motor to the finger module 6. Among them, the finger module 6 in this embodiment is divided into index finger, middle finger and ring finger modules, and each module adopts an equal-proportion structure. The little finger module is adaptively reduced in design on the premise of maintaining the same transmission topology. The thumb module is not integrated in the current embodiment due to significant differences in degrees of freedom of movement.

[0063] See Figure 4 , each finger module (6) includes a middle phalanx fixing plate 602, a proximal phalanx top plate 603, a proximal phalanx bottom plate 606, bilateral proximal phalanx side links 605 and a drive shaft fixing plate 607. One end of the drive shaft fixing plate 607 is connected to the palm fixing plate 5, and the other end is hinged to the proximal phalanx bottom plate 606 through a rotating shaft. The proximal phalanx bottom plate 606 and the proximal phalanx top plate 603 are connected to each other through the bilateral proximal phalanx side links 605 to form a parallelogram link mechanism; one end of the proximal phalanx top plate 603 is hinged to the middle phalanx fixing plate 602 through a rotating shaft, and a magic tape is provided on the middle phalanx fixing plate 602 for transmitting motion to the proximal phalanx and / or middle phalanx of the finger;

[0064] Both ends of the bilateral proximal phalanx side links are respectively hinged to the proximal phalanx top plate 603 and the proximal phalanx bottom plate 606 to realize the parallel displacement of the proximal phalanx top plate or the phalanx bottom plate. The rotational motion of the proximal phalanx bottom plate 606 is converted into the synchronous translational motion of the proximal phalanx top plate 603 through the parallelogram mechanism.

[0065] An angle sensor is also integrated on the proximal phalanx bottom plate for real-time detection of the joint angle. A fixed retaining ring is also provided on the proximal phalanx top plate to limit the rotation angle range of the bilateral proximal phalanx side links to -30° to 40°, preventing joint overload;

[0066] The proximal phalanx bottom plate is connected to the shaft sleeve 4 through a Bowden wire, and the torque output by the servo motor 1 is transmitted to the proximal phalanx bottom plate of the finger module 6 through the Bowden wire.

[0067] (2) The dynamics module is used to construct a simplified dynamics model of the finger module 6; specifically:

[0068] The exoskeleton robotic finger can swing along the vertical axis, but its swing angle is uncontrolled. For the convenience of analysis, the swing angle of the exoskeleton robotic hand is ignored, and the proximal, middle and distal phalanges of the exoskeleton robotic finger are regarded as a whole respectively. Then the exoskeleton robotic finger module can be simplified to Figure 5The shown three-link structure regards the proximal phalanx, middle phalanx, and distal phalanx as three connected link structures. The nodes from left to right are the metacarpophalangeal joint, proximal interphalangeal joint, and distal interphalangeal joint in sequence. A right-handed Cartesian coordinate system is established at the joints and the end respectively.

[0069] Based on the Lagrange equation Construct a simplified dynamic model, where represents the differentiation with respect to time, and τ i is the torque provided by the joint driver of the finger module, L is the Lagrangian function, and q i is the rotation angle of the i-th joint. Therefore, the establishment of the simplified dynamic model is transformed into the solution of the kinetic energy and potential energy of the upper limb finger module. The specific steps are as follows:

[0070] Kinetic energy calculation: First, the kinetic energy of the finger module of the multi-link mechanism can be regarded as the sum of the kinetic energies of each link

[0071]

[0072] Among them, T i (0 < i ≤ 3) is the kinetic energy of link i, m i is the mass of link i, I i is the inertia matrix of link i, v i and w i are the linear velocity vector and angular velocity vector at the end of link i respectively. By adding the kinetic energies of the three links in each finger module, the total kinetic energy of the hand exoskeleton mechanical module can be finally obtained.

[0073] Potential energy calculation: Considering that the potential energy received by the finger module is mainly gravitational potential energy and other potential energies can be ignored, the total potential energy of this finger module can be expressed in the following form

[0074]

[0075] Among them, m i is the mass of link i, g is the gravitational coefficient matrix, and y i is the vertical height of the center of mass of each link. Since y i is a function of the joint angle θ, the potential energy V of the final finger module can be described as a scalar function related to the joint angle θ.

[0076] Substituting the above formula into the Euler-Lagrange equation, we can get:

[0077]

[0078] In the above formula, M(θ) is the inertia matrix of the finger module, $\mathbf{C}$ is the matrix of Coriolis force and centrifugal force terms for the finger module, $\mathbf{G}(\theta)$ is the matrix of gravity terms, and $\tau$ is the matrix of torques required for the finger module to complete the desired trajectory. The above is the simplified dynamic model. Since the hand exoskeleton control system provided in this embodiment is a human-machine coupling system, the interaction force between humans and machines also needs to be considered. Therefore, the above dynamic model can be rewritten as:

[0079]

[0080] where $\tau$ inf is the control torque matrix provided by the joint driver (servo motor 1) of the finger module, $\tau$ int is the human-machine interaction torque matrix, is the joint friction matrix of the finger module, and $\tau$ d is the time-varying bounded external disturbance of the finger module.

[0081] (3) The outer-loop admittance controller is used to adaptively adjust the equivalent stiffness and damping parameters of the outer-loop admittance controller based on the deviation between the actual human-machine interaction force and the ideal human-machine interaction force, and combine the inertia and damping characteristics in the simplified dynamic model to generate a compliant trajectory compensation amount that conforms to the actual dynamic response; specifically:

[0082] In this embodiment, a control method combining inner and outer loops is provided for the hand exoskeleton control system. To achieve the goal of on-demand assistive control, the control block diagrams of the inner and outer loop control links are designed as Figure 6 shown. The entire control system is divided into an inner loop and an outer loop. The outer loop is responsible for evaluating the patient's motor ability and calculating the optimal assistive force. After converting the assistive force into the position difference between the finger module and the patient through the outer-loop admittance controller, the desired position of the finger module is transmitted to the inner loop, and the inner loop achieves accurate trajectory tracking through a certain trajectory tracking method.

[0083] Due to the environmental modeling error in the control system, the actual human-machine interaction force cannot reach the value of the ideal human-machine interaction force. In this way, the deviation between the two forces can be obtained by collecting and comparing through the human-machine interaction module. The deviation between the ideal and actual interaction forces will obtain an angle compensation amount through the outer-loop admittance controller. Inputting the angle compensation amount into the inner-loop position controller can achieve the effect of human-machine contact compliance. The meanings of the parameters of the outer-loop admittance controller are as follows: is the ideal joint angle, velocity, and acceleration, is the correction amount of the outer-loop admittance controller for the ideal joint angle, velocity, and acceleration, is the joint angle, velocity, and acceleration obtained after system compensation, and $\mathbf{f}$ d is the ideal human-machine interaction force, $\mathbf{f}$ e is the actual human-machine interaction force, and $\mathbf{M}$ d is the desired target inertia, $\mathbf{B}$d is the desired target damping, K d is the desired target stiffness, and e f is the interaction force error between the user and the finger module. Then the outer - loop admittance control method can be expressed by the following equation:

[0084]

[0085] By performing Laplace transform on the above equation, it can be transformed to the frequency - domain space for analysis, and we get:

[0086]

[0087] Δθ(s) is the function of the ideal joint angle in the frequency - domain space. After being corrected by admittance control, the desired position is:

[0088] θ c = θ d -Δθ(7)

[0089] For the piano - playing teaching scenario, the ideal human - machine interaction force f d = 0, which means that in this case, there is no interaction force between the human and the finger module, and the finger - playing trajectory conforms well to the expected trajectory, and the movement is completely dominated by the user, and the finger module only follows. When the user's playing effect is not good, there is a large conflict between the finger trajectory and the finger module, and an interaction force error e f occurs between the user and the finger module. When it exceeds a certain threshold, the finger module needs to actively provide auxiliary force to dynamically adjust the trajectory to achieve the effect of on - demand assistance.

[0090] (4) The inner - loop position controller is used to take the compliance trajectory compensation amount as the tracking target. By fusing the inverse - dynamics calculation of the simplified dynamics model, a composite control law combining the feed - forward compensation term and the sliding - mode surface function is constructed, and the output torque of the servo motor is dynamically adjusted based on the exponential reaching - law function to reduce the sliding - mode chattering and achieve finger - playing trajectory tracking; specifically:

[0091] The design of the outer - loop admittance controller has been completed above. The inner - loop uses the simplified dynamics model between the finger module and the human body constructed above, and a sliding - mode control is used to design the inner - loop position controller. According to the hand exoskeleton dynamics equation, the state variable x1 = θ c , The matrices of the dynamics model are rewritten as M(x1), C(x1,x2), G(x1), and the state - space expression of the hand exoskeleton mechanical module is obtained as:

[0092]

[0093] Among them:

[0094] ρ = ΔM(x1) + ΔC(x1, x2) + ΔG(x1) + F(x2) + τ d (9)

[0095] Where: ρ is the lumped modeling error of the dynamic simplified model, that is, the total uncertainty term. ΔM(x1), ΔC(x1, x2), and ΔG(x1) are the uncertainty terms of the dynamic model, and F(x2) is the corrected joint friction force matrix. The state-space expression of the mechanical module of the hand exoskeleton derived from the dynamic equation can clearly describe its dynamic characteristics and provide a mathematical basis for the design of the inner-loop position controller.

[0096] From the previous definitions, the joint angle tracking error e, the joint angular velocity tracking error and the joint angular acceleration tracking error are respectively expressed as:

[0097]

[0098] The sliding mode surface s is selected as a linear sliding mode surface, that is:

[0099]

[0100] Where Λ is the parameter matrix to be designed. By taking the derivative of the equation and substituting equations (8) and (10), we can obtain:

[0101]

[0102] M, C, and G are the abbreviations of the dynamic matrices. Let in the equation, and we can obtain the equivalent control law function of the finger module:

[0103]

[0104] The reaching law of the inner-loop position controller takes the exponential reaching law function, that is:

[0105]

[0106] k and ε are the reaching rate parameters. When k and ε are properly selected, it can ensure that the hand exoskeleton control system approaches stability, and the approaching speed of the system is mainly related to the value of k. Define the Lyapunov function as V = 1 / 2s 2 , take the derivative of it and substitute the exponential reaching law, we can obtain:

[0107]

[0108] The reason for adding the constant velocity approaching term to the exponential reaching law is as follows: It can be seen from the solution expression of Equation (14) that when using pure exponential reaching, the process of the moving point approaching the switching surface is asymptotic, the time for the system to reach the sliding mode surface is not finite, and the approaching velocity when reaching the sliding mode surface is very small, even tending to zero. After adding the constant velocity approaching term, the approaching velocity of the moving point approaching the switching surface is a constant value ε, ensuring that the system state vector can approach the equilibrium point at a relatively small velocity when reaching the sliding mode surface.

[0109] In the actual application process, due to the existence of the switching term, there will be obvious chattering in the sliding mode control, so the boundary layer thickness is introduced Use the saturation function to replace the sign function and reduce high-frequency switching:

[0110]

[0111] In summary, the final form of the control law of the hand exoskeleton control system is:

[0112]

[0113] (5) The online compensation unit is used to perform real-time estimation and feedforward compensation on the nonlinear disturbance term existing in the tracking error of the inner-loop position controller based on the RBF neural network.

[0114] Specifically:

[0115] This part is used to estimate the uncertain term ρ in the control law, and utilize the characteristic that the adaptive RBF neural network can approximate any nonlinear function with arbitrary precision in a compact set to predict the uncertain function of the hand exoskeleton system, and use its estimated value as the input of the feedforward end to the control system. The main structure of the RBF neural network includes an input layer, a hidden layer, and an output layer, as Figure 7 shown:

[0116] The shown RBF neural network includes n input layers, m hidden layers, and 1 output layer. In this neural network, use X = [x1, x2... x n T to represent the input layer data, and different dimensions and parameters can be selected according to different situations in actual applications; use h = [h1, h2... h m T to represent the output of the hidden layer. Each element represents a neuron, and the value of each neuron is calculated using the Gaussian basis function, which can be expressed as:

[0117]

[0118] In the formula, c j represents the center of the radial basis function; b j ​​represents the width of the radial basis function, \(j = 1, 2,\cdots\), representing the hidden layer nodes of the RBF neural network. The output of the RBF neural network is the linear sum of the hidden layer nodes, taking as the input vector, that is:

[0119]

[0120] where: is the estimated value of the ideal weight of the output layer, defined as represents the weight error, then the estimation error of the total uncertainty term of the inner loop position controller is:

[0121]

[0122] where: \(\gamma\) is the estimation error of the uncertainty term of the dynamic simplified model. In order to achieve the adaptive adjustment of the neural network weights while satisfying the system stability, the adaptive law of the neural network weights is taken as:

[0123]

[0124] where: \(\Gamma\) is a positive diagonal matrix to be designed. Thus, the control link of the inner loop position controller is improved, that is, the design of the entire closed-loop control system of the hand exoskeleton system is completed.

[0125] Preferably, in the piano playing teaching environment, due to the flexibility of finger playing, which brings system complexity, the effect of using a controller with fixed admittance parameters is limited. Therefore, an adaptive admittance control for the hand exoskeleton is introduced, and fuzzy rules are used to achieve the adaptive adjustment of admittance parameters, and the stiffness and damping parameters can be adjusted in real time according to the system input. For exoskeleton control, the system needs to ensure stability while optimizing the interaction compliance. Therefore, the trajectory error and the human-machine interaction force are used as the input variables of the fuzzy controller, and specific fuzzy rules can be designed according to experience and control objectives.

[0126] Design rules according to the combined relationship between the trajectory error and the interaction force:

[0127] 1. High error and low interaction force: The system needs to provide more auxiliary force to correct finger movements, but no damping needs to be increased;

[0128] 2. High error and high interaction force: The user is already applying a large external force, indicating that more compliance is needed, so damping needs to be increased;

[0129] 3. Low error and low interaction force: The system state is close to ideal, and the stiffness and damping are reduced to increase compliance;

[0130] 4. Low error and high interaction force: Although the error is small, the user applies a large force, and there may be a change in intention, so the damping needs to be moderately increased;

[0131] 5. Medium error and medium interaction force: Provide moderate auxiliary force and damping to ensure motion stability;

[0132] The fuzzy inference adopts the Mamdani algorithm, and the Gaussian function is taken as the membership function to divide the trajectory tracking error e into fuzzy sets {Negative Big (NB), Negative Small (NS), Zero (ZE), Positive Small (PS), Positive Big (PB)}; the human-machine interaction force F ext is divided into fuzzy sets {Low, Medium, High}. The fuzzy adjustment rules are shown in Table 1:

[0133] Table 1 Fuzzy adjustment rule table

[0134]

[0135] Finally, in order to verify the effectiveness of the proposed method, this embodiment experimentally verified the control effect of the control system of the present invention described in formula (17). The simulation time was selected as the time of a piano-playing action, 3 s. The exoskeleton desired trajectory was the high finger-lifting action of piano-playing: α(t) = sin(2πt / 0.7) - 0.88, with a period of 0.7 s. In order to illustrate the robustness and superiority of the proposed method in control, under external disturbance conditions, tracking error comparison diagrams of different control methods, including PID control, sliding mode control, and sliding mode compensated by RBF network, were made. In addition, schematic diagrams of the auxiliary force change of the on-demand auxiliary control were also given to verify the control effect of the on-demand auxiliary. Among them, the inner-loop control parameters of the on-demand auxiliary control method of the present invention were set as: k = 40; Λ = 80; ε = 0.3, time delay T = 0.001 s, and the RBF neural network control parameters were set as: b j = 5, the number of hidden layer nodes was 20, c j was uniformly distributed in the interval [-1, 1], and Γ was the identity matrix.

[0136] Figure 8 is the trajectory tracking diagram of this control method. It can be seen that the trajectory of the joint movement during piano-playing is well tracked. Figure 9 is the error change diagram of the joint trajectory tracking control effect of the control system and method of the present invention and other methods, Figure 10 is the estimated value change diagram of the RBF neural network for compensating the unknown term and disturbance term, Figure 11 is the schematic diagram of the auxiliary force change of the on-demand auxiliary control. It can be seen from the experimental results that the control system and method proposed by the present invention not only have good robustness, but also compared with traditional self-control methods, as the control operation time goes deeper, the joint trajectory tracking error becomes smaller and smaller, and the on-demand auxiliary control can provide auxiliary torque when there is a trajectory deviation between the human hand and the exoskeleton.

[0137] In summary, piano playing, as a professional hand motor skill, has strict requirements for the posture of the human hand, as well as the movement and force of key pressing. Especially for beginners, developing good playing habits in the early stage is crucial for future learning and development. Aiming at the problems of poor movement standardization and lack of feedback in autonomous practice during the early stage of piano learning, the present invention proposes an intelligent exoskeleton control system and method for the hand that integrates sliding mode control technology and fuzzy adaptive admittance control algorithm. The system realizes the coupled movement of three joints (metacarpophalangeal joint MCP, proximal interphalangeal joint PIP, and distal interphalangeal joint DIP) through an underactuated Bowden cable drive mechanism. It innovatively uses sliding mode position control combined with RBF neural network technology to improve the robustness of trajectory tracking, and dynamically adjusts the human-machine interaction impedance parameters through fuzzy adaptive admittance control, constructing an intelligent control architecture for on-demand assistance.

[0138] Specifically, the hand exoskeleton control system uses a sliding mode inner-loop position controller to eliminate the trajectory deviation caused by transmission hysteresis; combines the RBF network to compensate for the deviation caused by system identification and external disturbances, improving the overall robustness. In addition, the system also realizes the control idea of on-demand assistance. The fuzzy adaptive admittance controller monitors the finger movement state in real time and applies an auxiliary torque only when the deviation between the joint angle and the target trajectory is too large, which not only ensures the training freedom but also maintains the movement standardization. In short, the designed piano-assisted teaching hand exoskeleton system effectively solves the problems of excessive rigidity of traditional teaching robots and interference with the user's proprioception, and can better meet the requirements in the initial stage of piano learning, helping users improve the training effect.

[0139] The present invention is not limited to the embodiments described above. The above description of the specific embodiments is intended to describe and illustrate the technical solutions of the present invention. The above specific embodiments are only illustrative and not restrictive. Without departing from the spirit of the present invention and the scope protected by the claims, those of ordinary skill in the art can make many specific transformations in various forms under the inspiration of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A hand exoskeleton control system for on-demand assisted teaching of the piano, characterized in that, It includes a hand exoskeleton mechanical module, a dynamics module, an outer loop admittance controller, an inner loop position controller, and an online compensation unit that are connected in sequence; The hand exoskeleton mechanical module consists of a servo motor (1), a dynamic torque sensor (2), a hand exoskeleton base (3), a bushing (4), a palm fixing plate (5), and a finger module (6); The servo motor (1) and the dynamic torque sensor (2) are fixed on the hand exoskeleton base (3), the palm fixing plate (5) is fixed on the upper surface of the hand exoskeleton base (3), the transmission shaft of the servo motor (1) is connected to the input shaft of the dynamic torque sensor (2) through a coupling, the output shaft of the dynamic torque sensor (2) is fixed to the bushing (4) through a flat key, a Bowden wire is fixed on the bushing (4), and the bushing (4) is connected to the finger module (6) through the Bowden wire to realize the transmission of the power of the servo motor to the finger module 6; The dynamics module is used to construct a dynamic simplified model of the finger module (6); The outer loop admittance controller is used to adaptively adjust the equivalent stiffness and damping parameters of the outer loop admittance controller based on the deviation between the actual human-machine interaction force and the ideal human-machine interaction force, and combine the inertia and damping characteristics in the dynamic simplified model to generate a compliant trajectory compensation amount that conforms to the actual dynamic response; The inner loop position controller is used to take the compliant trajectory compensation amount as the tracking target, construct a composite control law that combines the feedforward compensation term and the sliding mode surface function through the inverse dynamics calculation of the dynamic simplified model, and dynamically adjust the output torque of the servo motor based on the exponential reaching law function to reduce the sliding mode chattering and realize the finger playing the piano trajectory tracking; The online compensation unit is used to perform real-time estimation and feedforward compensation on the non-linear disturbance term existing in the tracking error of the inner loop position controller based on the RBF neural network.

2. The hand exoskeleton control system according to claim 1, wherein The finger module (6) includes a middle phalanx fixing plate, a proximal phalanx top plate, a proximal phalanx bottom plate, bilateral proximal phalanx side links, and a drive shaft fixing plate, One end of the drive shaft fixing plate is connected to the palm fixing plate, and the other end is hinged to the proximal phalanx bottom plate through a rotating shaft. The proximal phalanx bottom plate and the proximal phalanx top plate are connected to each other through bilateral proximal phalanx side links to form a parallelogram link mechanism. One end of the proximal phalanx top plate is hinged to the middle phalanx fixing plate through a rotating shaft, and a magic tape is provided on the middle phalanx fixing plate for transmitting motion to the finger proximal phalanx and / or the middle phalanx; Both ends of the bilateral proximal phalanx side links are respectively hinged to the proximal phalanx top plate and the proximal phalanx bottom plate to realize the parallel displacement of the proximal phalanx top plate or the phalanx bottom plate; An angle sensor is also integrated on the proximal phalanx bottom plate for real-time detection of the joint angle, and a fixed retaining ring is also provided on the proximal phalanx top plate to limit the rotation angle range of the bilateral proximal phalanx side links to -30° - 40° to prevent joint overload; The proximal phalanx bottom plate is connected to the bushing (4) through a Bowden wire, and the torque output by the servo motor (1) is transmitted to the proximal phalanx bottom plate of the finger module (6) through the Bowden wire.

3. The hand exoskeleton control system according to claim 1, characterized in that, The outer - loop admittance controller realizes the adaptive adjustment of admittance parameters by using fuzzy rules. Taking the trajectory error and the human - machine interaction force as input variables, it outputs an online adjustment strategy for admittance parameters. The fuzzy rules adopt the Mamdani algorithm, and the membership function is a Gaussian function. The admittance parameters include equivalent stiffness and damping.

4. The hand exoskeleton control system according to claim 1, wherein The RBF neural network has several input layers, several hidden layers, and one output layer. It updates the weights of the hidden layer online through an adaptive law to approximate the uncertainties in the inner - loop position control.

5. A method for on-demand assisted control of a hand exoskeleton, based on the hand exoskeleton control system according to any one of claims 1-4, characterized in that, It includes: The human - machine interaction force is collected in real - time through a dynamic torque sensor and a double - threshold judgment is made with a preset ideal interaction force. When the interaction - force error exceeds the threshold, the outer - loop admittance controller is triggered to dynamically adjust the stiffness and damping parameters of the controller and generate a compliant - trajectory compensation amount. The compliant - trajectory compensation amount is input into the inner - loop position controller, and the required driving torque is calculated and output according to the sliding - mode surface function and the exponential reaching - law function. The RBF neural network is used to estimate the non - linear disturbance term in the position - tracking error online and the estimated value is used as a feed - forward compensation term. The above steps are continuously and cyclically executed to achieve the piano - playing assistance that can be assisted as needed.

Images