An intelligent control method for upper limb rehabilitation robot

Abstract

This invention provides an intelligent control method and system for an upper limb rehabilitation robot. Addressing the uncertainty of the discrete-time dynamic model of a real robotic arm, this invention designs a tracking error and then constructs an adaptive neural network controller based on discrete deterministic learning theory. This controller can accurately model / learn the internal unknown dynamics along a periodic trajectory, and then utilize the learned knowledge to construct an experience-based learning controller. Furthermore, by combining interpersonal skill transfer methods, the control performance of the rehabilitation robot in uncertain environments is improved. This invention achieves rapid convergence, high precision, and better transient performance, which is of great significance for improving the efficiency of rehabilitation training with upper limb rehabilitation robots.

Description

Technical Field

[0001] This invention belongs to the field of robot control technology and relates to an intelligent control method for an upper limb rehabilitation robot. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Upper limb motor dysfunction is a common phenomenon following aging, stroke, sports injuries, or spinal cord injuries. According to neurorehabilitation theory, repetitive and task-oriented training plays a positive role in rebuilding upper limb nerve and muscle function. Addressing the shortcomings of traditional therapist-assisted rehabilitation, such as high labor intensity, long treatment time, high cost, poor continuity, and poor repeatability, robot-assisted rehabilitation therapy has been proven to be an effective treatment method for promoting neuroplasticity and motor function recovery in patients.

[0004] Control strategy is one of the core components of upper limb rehabilitation robots. However, most upper limb rehabilitation robots currently employ relatively crude control strategies, which are unable to learn human behavior and have significant limitations in control accuracy and transient performance in uncertain discrete-time system environments. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes an intelligent control method for upper limb rehabilitation robots. This method is based on Discrete-time Deterministic Learning (DTDL) theory and consists of two parts: an Adaptive Neural Network Controller (ANNC) and an Experience-Based Learning Controller (LC), used for knowledge acquisition and utilization, respectively. This invention addresses the uncertainty of the discrete-time dynamic model of a practical robotic arm by designing a tracking error, then constructing a suitable ANNC that satisfies persistent excitation (PE), and combining it with the Human-robot Skill Transfer (HRST) method. This allows for precise modeling / learning of the internal unknown dynamics along a periodic trajectory, and the learned knowledge is then used to construct a LC-based control system, improving the control performance of the rehabilitation robot in uncertain environments. This invention achieves rapid convergence, high precision, and better transient performance, which is of great significance for improving the efficiency of upper limb rehabilitation robot rehabilitation training.

[0006] According to some embodiments, the present invention adopts the following technical solution:

[0007] A method for intelligent control of an upper limb rehabilitation robot includes the following steps:

[0008] Acquire several teaching trajectories, which are obtained from customized rehabilitation training trajectories based on rehabilitation needs;

[0009] Based on the teaching trajectory, a periodic reference trajectory is determined, and motion representation and skill modeling are performed on each periodic reference trajectory to fit the final reference trajectory.

[0010] A discrete-time adaptive radial basis function neural network controller is constructed to learn the unknown dynamics in the upper limb rehabilitation robot system during the tracking control process. The interaction forces between the rehabilitation client and the upper limb rehabilitation robot system are approximated / learned. The learned knowledge is used to construct an experience-based learning controller and further improve the control performance.

[0011] As an alternative implementation method, the specific process of customizing rehabilitation training trajectories according to rehabilitation needs includes: determining corresponding rehabilitation training movements based on the actual needs of the person requiring rehabilitation, and obtaining multiple periodic reference trajectories through repeated demonstrations by the instructor.

[0012] As an alternative implementation method, the specific process of performing motion representation and skill modeling on each periodic reference trajectory to fit the final reference trajectory includes:

[0013] For inconsistent time lengths in the teaching trajectories, spline interpolation algorithm is used to fill in the gaps.

[0014] If the starting positions of the teaching trajectories are inconsistent, the generalized time warping algorithm is used to align the multiple teaching trajectories.

[0015] The Gaussian mixture model and Gaussian mixture regression are used to integrate the multiple teaching trajectories after completion and alignment into a final reference trajectory.

[0016] Furthermore, when using a Gaussian mixture model, the parameters used in the model are the model parameters of the Gaussian mixture model obtained by the expectation-maximization algorithm.

[0017] As an alternative implementation, the tracking learning process of the discrete-time adaptive radial basis function neural network controller includes:

[0018] Obtain the discrete-time dynamic model of the upper limb rehabilitation robot;

[0019] The discrete-time dynamic model is transferred to the output feedback control system to establish a dynamic model in the form of a standard system.

[0020] Using the backstepping method, the tracking error is defined, and the ideal form of the controller is obtained through the dynamic equation of the system error;

[0021] An adaptive radial basis function neural network controller is obtained by approximating / learning the unknown dynamics of the robot and the uncertainties arising from the interaction between the rehabilitation client and the upper limb rehabilitation robot system using radial basis function neural networks.

[0022] The stability analysis and weight update of the discrete-time adaptive radial basis function neural network controller are performed based on Lyapunov stability theory.

[0023] Furthermore, to avoid interference from the system's affine terms on learning, the learning error is converted into a discrete linear time-varying perturbation through state transition, and closed-loop learning is performed.

[0024] As an alternative implementation, the process of building an experience-based learning controller includes:

[0025] The neural network weights after stable convergence during the tracking control process of the discrete-time adaptive radial basis function neural network controller are saved as constant values.

[0026] An experience-based learning controller is constructed using a constant neural network to further improve control performance in the same or similar tasks.

[0027] An intelligent control system for an upper limb rehabilitation robot includes:

[0028] The fitting module is configured to determine a periodic reference trajectory based on the acquired teaching trajectory, perform motion representation and skill modeling on each periodic reference trajectory, and fit the final reference trajectory.

[0029] The training module is configured as a discrete-time adaptive radial basis function neural network controller to identify unknown dynamics in the upper limb rehabilitation robot system during the tracking control process, and to approximate / learn and store the interaction forces between the rehabilitation client and the upper limb rehabilitation robot system.

[0030] The learning control module is configured as an experience-based learning controller, which further improves control performance by utilizing knowledge learned from the adaptive radial basis function neural network controller.

[0031] A computer-readable storage medium storing a plurality of instructions adapted for loading by a processor of a terminal device and executing steps in the method.

[0032] A terminal device includes a processor and a computer-readable storage medium, the processor being configured to implement instructions; the computer-readable storage medium being configured to store a plurality of instructions adapted to be loaded by the processor and executed in accordance with the steps of the method.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] Compared with the existing control strategies for passive training of upper limb rehabilitation robots, the control method proposed in this invention not only considers the discrete-time characteristics of the actual upper limb rehabilitation robot, but also achieves higher tracking accuracy and better transient performance in the neighborhood where the tracking error of any periodic trajectory in passive training eventually tends to zero.

[0035] The training trajectory in this invention is based on the patient's needs and is optimized and integrated through multiple drag-and-drop demonstrations on the patient's healthy side, fully taking into account the patient's needs and the subtle differences between multiple human movements.

[0036] The method in this invention transforms a learning error-free system into a discrete linear time-varying perturbation system through state transition, thus solving the problem of learning failure due to the presence of uncertain affine terms in the discrete system of the upper limb rehabilitation robot.

[0037] The method in this invention can accurately model unknown dynamics and unpredictable disturbances in nonlinear systems locally, and store the learned experience knowledge in the form of a constant neural network. For subsequent identical or similar control tasks, the stored knowledge can be directly called for control, eliminating the need for online calculation of controller parameters, saving energy and control time in the rehabilitation training system, and further improving the dynamic control performance in transient processes.

[0038] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0039] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0040] Figure 1 This is an overall control block diagram of the upper limb rehabilitation robot in this invention;

[0041] Figure 2 This is an overall block diagram of the HRST technology in this invention;

[0042] Figure 3 This is a flowchart illustrating the specific implementation of the intelligent control method for upper limb rehabilitation robots based on DTDL and HRST in this invention.

[0043] Figure 4 This is a schematic diagram illustrating the actual implementation of the intelligent control method for upper limb rehabilitation robots based on DTDL and HRST in this invention. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0045] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0046] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0047] An intelligent control method for an upper limb rehabilitation robot addresses the uncertainties in the discrete-time dynamic model of the actual robotic arm and the human-machine interaction process. First, it constructs an adaptive neural network controller (ANNC) using backstepping and Discrete-time Deterministic Learning (DTDL) theory. This ANNC can accurately model / learn the unknown internal dynamics along a periodic trajectory. Then, it uses the learned knowledge to construct an experience-based learning controller (LC). The control method designed in this invention achieves fast convergence, high precision, and better transient performance. Furthermore, this method can directly achieve rapid control for subsequent identical or similar control tasks using the learned knowledge. Because it utilizes the learned knowledge, the controller parameters do not need to be calculated online, shortening the control time and saving system energy.

[0048] This invention also addresses the problem that current rehabilitation robots often use pre-set trajectories with low personalization. It combines a DTDL-based intelligent control method for rehabilitation robots with a Human-robot Skill Transfer (HRST) method. The basic idea is to define teaching and training modes based on patient needs and daily life requirements, then use HRST to transmit personalized movement patterns to the rehabilitation robot. Finally, it utilizes a control method based on Discrete Deterministic Learning Theory (DDL) to improve the robot's control performance in uncertain environments. This intelligent control method is of great significance for improving the efficiency of upper limb rehabilitation robot training.

[0049] First, to facilitate understanding by technical personnel, the key technical points involved in this invention will be introduced:

[0050] The first part is the construction and learning of the controller.

[0051] This mainly includes designing ANNC and LC based on the discrete-time dynamic model of the upper limb rehabilitation robot and DTDL theory.

[0052] Based on the following discrete-time dynamic model of the upper limb rehabilitation robot:

[0053]

[0054] Where k is a discrete time point, and T is a discrete time interval; j k ,v k J represents joint position and joint velocity, respectively. k =j k +T s v k M(j) k ) and M(J k F(j) is the inertia matrix; k ,v k ) is the Coriolis force-centrifugal force and gravitational torque matrix; τ k It controls the input torque; It is the torque for human-computer interaction.

[0055] The discrete dynamics of the above-mentioned upper limb rehabilitation robot are transmitted to the output feedback control system to establish a dynamic model in the form of a standard system.

[0056]

[0057] in,

[0058] U(X k ) = M -1 (J k ), X k =[j k ,v k ] T .

[0059] Design error variables:

[0060]

[0061] Among them, z1 k z2 k For joint position and velocity tracking errors, For periodic or quasi-periodic joint position reference trajectory; θ k It is a virtual controller, in the form of:

[0062]

[0063] In the formula, p is the designed controller gain constant.

[0064] According to formulas (2) and (3), we can obtain:

[0065]

[0066] To ensure the stability of the closed-loop system, the required control input is selected as follows:

[0067]

[0068] In the formula, m is the designed controller gain constant. This refers to the unknown dynamics within the upper limb rehabilitation robot system.

[0069] By appropriately designing a discrete-time adaptive neural network controller based on a discrete upper limb rehabilitation robot dynamics model and DTDL, a method is used to accurately identify (learn) unknown dynamics in the upper limb rehabilitation robot system during tracking control. Specifically:

[0070]

[0071] The neural network used is a radial basis function neural network (RBFNN). For RBFNN output, Let φ(Y) be the weight of the RBFNN. k ) is the set of radial basis functions. This serves as the input for the RBFNN. Unlike traditional RBFNNs, the RBFNN in this invention requires persistent excitation (PE) to achieve learning.

[0072] Based on Lyapunov stability theory and DTDL theory, the neural network weight update law is designed as follows:

[0073]

[0074] Where, Γ=Γ T >0 is a positive definite diagonal matrix, and σ>0 is a constant.

[0075] Learning closed-loop systems from adaptive RBFNN;

[0076] According to DTDL theory, when the neurons of an RBFNN along a periodic training trajectory satisfy the PE condition, both the state error and weight estimation are bounded and converge exponentially. However, due to the uncertain affine term U(X) in the upper limb rehabilitation robot system... k ) = M-1 (J k The presence of [something] amplifies the error, leading to a failure to learn. To address this issue, this patent transforms the learning error system into a discrete linear time-varying (DLTV) perturbation system through state transition, achieving exponential stability and thus a learning effect. Specifically: Let [something]... in and The DLTV system can then be represented as:

[0077]

[0078] In the formula, D k =T 2 Γφ(Y k )U(X k )φ T (Y k ), δ ζ Let ζ represent the approximation error of the neural network, and let the subscript ζ indicate that the neurons of the RBFNN satisfy the PE condition.

[0079] We design an experience-based learning controller using the learned knowledge; based on DTDL theory, we adjust the weights of the stable and converged neural network. Save as constant value Specifically:

[0080]

[0081] Among them, [k α ,k β [ ] represents the time interval after the system has reached stable convergence.

[0082] Then, the obtained constant neural network controller is used. The controller is modified, and the controller takes the following form:

[0083]

[0084] For the same control task, experience-based learning controllers have better transient performance and higher control accuracy. Moreover, since online calculation of controller parameters is no longer required, system energy and control time are saved, which is of great value for rehabilitation training.

[0085] The second part is custom trajectory reproduction based on HRST technology, such as... Figure 2 As shown, it mainly includes the following steps:

[0086] Step (1): Acquiring teaching data;

[0087] Based on the patient's actual needs, select appropriate rehabilitation training movements; obtain multiple periodic reference trajectories through repeated demonstrations by the instructor;

[0088] Step (2): Representation and modeling of motor skills;

[0089] After the teaching is completed, motion representation and skill modeling are required for the acquired trajectory data to ultimately fit a reference trajectory rich in patient individuality. In this process, the issue of trajectories alignment must be considered. Even when the same person performs multiple repetitive actions, the duration and starting position of the teaching trajectory will not be consistent. To address the inconsistency in duration, spline interpolation is used for completion. To address the inconsistency in starting position, Generalized Time Warping (GTW) is used to align the multiple teaching trajectories. Consider the time sequence {U1,…,U...} of m teaching trajectories. m The GTW cost minimization function is:

[0090]

[0091] In the formula, W i and V i These are nonlinear time transformation and low-dimensional spatial embedding, respectively, φ(V i )and It is a regularization function. For each U i GTW can find a W i and V i So that sequence V i T U i W i It aligns well with other sequences in the least-squares sense. To optimize the cost function, GTW uses a Gauss-Newton algorithm with linear complexity for sequence length to optimize the time warp function, employs multi-set canonical correlation analysis to account for dimensionality differences, and uses a more flexible warp model parameterized by a set of monotonic bases to compensate for temporal and spatial variations.

[0092] By using Gaussian mixture models and Gaussian mixture regression, the multiple teaching trajectories are integrated into a final reference trajectory. The final control variables can be expressed as:

[0093]

[0094] Among them, h i (x) are the normalized weights, and the parameters in the above formula are the model parameters of the Gaussian mixture model obtained by the expectation-maximization algorithm.

[0095] Step (3): Skill transfer.

[0096] After acquiring the skill characteristics, the learned motion strategy control variables can be mapped to the controller of the robotic arm. The robot can then reproduce the motion skills of the patient's healthy side to complete the rehabilitation training of the affected side.

[0097] In some embodiments, during the task reproduction stage, it is necessary to select a suitable upper limb rehabilitation robot controller, which can be the intelligent control method based on discrete deterministic learning introduced in the previous section.

[0098] As a specific application, such as Figure 3 and Figure 4 As shown, it includes:

[0099] 1) First, a rehabilitation training trajectory is customized according to the patient's needs, and the patient demonstrates it multiple times on the healthy side (if the patient's healthy side cannot complete the demonstration, the therapist will complete it on their behalf). After integration and optimization by the algorithm in Part 2, the final rehabilitation training reference trajectory is obtained.

[0100] 2) Use an upper limb rehabilitation robot with an embedded ANNC controller to drive the patient to move, and use neural networks to approximate / learn the internal dynamics of the upper limb rehabilitation robot and the interaction force between the patient and the robot.

[0101] 3) Utilize the knowledge learned in 2) to construct an experience-based learning controller. When patients perform the same or similar rehabilitation training tasks in the future, the experience-based learning controller can work quickly and further improve control performance.

[0102] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.

[0103] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 illustrations and / or block diagrams. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0104] 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.

[0105] 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.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0107] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. An intelligent control method for an upper limb rehabilitation robot, characterized in that, Includes the following steps: Acquire several teaching trajectories, which are obtained from customized rehabilitation training trajectories based on rehabilitation needs; Based on the teaching trajectory, a periodic reference trajectory is determined, and motion representation and skill modeling are performed on each periodic reference trajectory to fit the final reference trajectory. The specific process includes: For inconsistent time lengths in the teaching trajectories, spline interpolation algorithm is used to fill in the gaps. If the starting positions of the teaching trajectories are inconsistent, the generalized time warping algorithm is used to align the multiple teaching trajectories. The multiple teaching trajectories after completion and alignment are integrated into a final reference trajectory using Gaussian mixture model and Gaussian mixture regression. When using a Gaussian mixture model, the parameters used in the model are the model parameters of the Gaussian mixture model obtained by the expectation-maximization algorithm. A discrete-time adaptive neural network controller is constructed to identify unknown dynamics in the upper limb rehabilitation robot system during the tracking control process, and to approximate / learn and store the interaction forces between the rehabilitation client and the upper limb rehabilitation robot system. By leveraging knowledge learned from discrete-time adaptive neural network controllers, an experience-based learning controller can be constructed to further improve control performance in the same or similar tasks. Specifically, this involves designing an adaptive neural network controller and a learning controller based on the discrete-time dynamic model and discrete deterministic learning theory of the upper limb rehabilitation robot. The discrete dynamics of the upper limb rehabilitation robot are transferred to the output feedback control system to establish a dynamic model of the standardized system form; A discrete-time adaptive neural network controller was designed using a discrete-time dynamic model of an upper limb rehabilitation robot and discrete deterministic learning theory. Discrete-time adaptive neural network controllers require the use of radial basis function neural networks, and the neuron vectors must satisfy the continuous excitation condition; Design a neural network weight update law based on Lyapunov stability theory and discrete deterministic learning theory; The learning process of a discrete-time adaptive neural network controller requires converting the learning error into a discrete linear time-varying disturbance through state transitions to overcome the interference of affine terms and perform closed-loop learning.

2. The intelligent control method for an upper limb rehabilitation robot as described in claim 1, characterized in that, The specific process of customizing rehabilitation training tracks according to rehabilitation needs includes: determining the corresponding rehabilitation training movements based on the actual needs of the person requiring rehabilitation, and obtaining multiple periodic reference tracks through repeated demonstrations by the instructor.

3. The intelligent control method for an upper limb rehabilitation robot as described in claim 1, characterized in that, is The weights of the neural network after learning and stable convergence are saved as constant values, and an experience-based learning controller is constructed using these constant values.

4. An intelligent control system for an upper limb rehabilitation robot, implementing the intelligent control method for an upper limb rehabilitation robot as described in any one of claims 1-3, characterized in that, include: The fitting module is configured to determine a periodic reference trajectory based on the acquired teaching trajectory, perform motion representation and skill modeling on each periodic reference trajectory, and fit the final reference trajectory. The training module is configured as a discrete-time adaptive radial basis function neural network controller to identify unknown dynamics in the upper limb rehabilitation robot system during the tracking control process, and to approximate / learn and store the interaction forces between the rehabilitation client and the upper limb rehabilitation robot system. The learning control module is configured as an experience-based learning controller, which further improves control performance by utilizing knowledge learned from the adaptive radial basis function neural network controller.

5. A computer-readable storage medium, characterized in that, It stores multiple instructions, which are adapted to be loaded by the processor of the terminal device and executed as steps in the method of any one of claims 1-3.

6. A terminal device, characterized in that, It includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store a plurality of instructions adapted to be loaded by the processor and executed as steps in the method of any one of claims 1-3.

Images