Upper limb rehabilitation robot discrete neural network control method and system

Abstract

The application discloses a kind of upper limb rehabilitation robot discrete neural network control method and system, the method includes: each joint module of upper limb rehabilitation robot is regarded as a subsystem, adaptive neural network controller is constructed based on discrete barrier lyapunov function;Determine weight update rate, according to discrete determination learning theory, experience-based learning controller is constructed;On the basis of experience-based learning controller, the fusion of knowledge is realized using least square method based on space constraint, and experience-based learning controller based on knowledge fusion is constructed;Variable admittance control model is constructed, and the output of variable admittance control model is used as the reference trajectory of experience-based learning controller based on knowledge fusion, and the training of corresponding rehabilitation action is executed according to the rehabilitation demand of user.The application realizes the intelligentization and safety high-performance control of upper limb rehabilitation robot in different individuals and different recovery periods.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of rehabilitation robot control, and particularly relates to a discrete neural network control method and system for an upper limb rehabilitation robot. BACKGROUND

[0002] The statements herein are provided only to complement the background of the present application and are not necessarily indicative of the prior art.

[0003] With the aggravation of population aging and the rise of stroke incidence, the rehabilitation needs of patients with upper limb motor dysfunction are growing. Traditional artificial rehabilitation treatment has problems such as shortage of therapist resources and difficulty in quantifying training intensity, prompting the rapid development of rehabilitation robot technology. Upper limb rehabilitation robots can effectively promote neural plasticity by providing precise assisted movement training, and have become an important research direction in the field of rehabilitation medicine.

[0004] The existing control methods for upper limb rehabilitation robots mainly face three technical challenges: (1) strong nonlinearity and time-varying characteristics of human-robot interaction system: factors such as changes in patient muscle tension and dynamic fluctuations in joint impedance make it difficult to accurately establish the system dynamics model; (2) contradiction between individual differences and training adaptability: different rehabilitation stages require differentiated assistance strategies, and traditional control methods have limitations in personalized adjustment and generalization ability; (3) strictness requirements of safety constraints: during rehabilitation training, it is necessary to ensure that parameters such as joint movement range and speed are always within the physiological safety threshold to avoid secondary injury.

[0005] In current mainstream control methods, PID control is simple in structure but difficult to handle nonlinear coupling; sliding mode control has strong robustness but is prone to high-frequency chattering; traditional adaptive control relies on accurate model prior knowledge. In recent years, neural networks have been introduced into rehabilitation robot control due to their strong nonlinear approximation ability. However, most adaptive neural network control methods only focus on the general function approximation ability of neural networks, ignoring the ability of neural networks to acquire, represent, and utilize knowledge in dynamic control processes with uncertainties. Therefore, even for the same or similar control tasks, control parameters must be updated online, resulting in poor control performance at the initial stage, increased computational burden, and time cost. Based on this, the deterministic learning theory is proposed, which realizes closed-loop learning of unknown dynamics of the system in the adaptive control process.

[0006] However, it is worth noting that existing robotic arm system control methods based on deterministic learning are all for single tasks, and their learned knowledge only covers a narrow area around the actual motion trajectory of the system in the state space, and cannot form knowledge representation in a larger area. Rehabilitation robots usually interact with patients with varying degrees of injury, and upper limb injuries often exhibit changing rather than constant states during rehabilitation.

[0007] In addition, unlike the passive control in the early stage of rehabilitation, the active participation of patients is required in the middle and late stages of rehabilitation to improve the rehabilitation effect through active control. The traditional deterministic learning lacks the generalization ability in variable working conditions and multi-task scenarios, and obviously cannot cope with such interactive control scenarios. Therefore, it is necessary to explore the closed-loop learning and control problem in a larger range to improve the effect of motor training of the rehabilitation robot facing different individuals and different recovery periods.

[0008] The barrier Lyapunov function technology provides a theoretical framework for the safety constraint control of the rehabilitation robot, and existing research has applied it to joint angle limit processing. However, the existing barrier Lyapunov function-based method is mainly designed for continuous-time systems, while the actual rehabilitation robot control system mostly uses a discretized digital controller. In the discretization process, the differential form of the continuous barrier Lyapunov function is difficult to directly convert into a difference equation, resulting in insufficient stability proof of the discrete system and a significant increase in the risk of constraint violation. Therefore, it is necessary to design a barrier Lyapunov function for a discrete system.

[0009] In summary, in order to cope with the challenges existing in the current upper limb rehabilitation robot, it is necessary to develop an advanced control strategy to realize the precise perception and precise control of the patient's motor ability by the rehabilitation robot under the premise of ensuring the safety of movement. This has important clinical value for improving the effect of rehabilitation training and shortening the rehabilitation period. SUMMARY

[0010] The purpose of the present application is to overcome the deficiencies in the prior art, and to provide a discrete neural network control method and system for an upper limb rehabilitation robot, which realizes intelligent and safe high-performance control of the upper limb rehabilitation robot for different individuals and different recovery periods.

[0011] In order to achieve the above-mentioned purpose, the present application is realized by the following technical solutions:

[0012] On the one hand, the technical scheme of the present application provides a discrete neural network control method for an upper limb rehabilitation robot, comprising:

[0013] Each joint module of the upper limb rehabilitation robot is regarded as a subsystem, and an adaptive neural network controller is constructed based on a discrete barrier Lyapunov function; the weight update rate is determined, and an experience-based learning controller is constructed based on the discrete deterministic learning theory;

[0014] On the basis of the experience-based learning controller, the knowledge fusion is realized by using the least square method based on spatial constraints, and an experience-based learning controller based on knowledge fusion is constructed;

[0015] A variable admittance control model is constructed, and an output of the variable admittance control model is taken as a reference trajectory of a learning controller based on knowledge fusion experience, so that the user can perform corresponding rehabilitation action training according to rehabilitation needs.

[0016] In at least one embodiment, each standard joint module in the upper limb rehabilitation robot is regarded as a subsystem, and is connected to each other through coupling torques of the whole upper limb rehabilitation robot system; the coupling torques are converted into a dynamic model in the form of a norm system with output feedback; a virtual controller and an ideal adaptive neural network controller are constructed by considering a symmetric potential barrier Lyapunov function.

[0017] In at least one embodiment, the ideal adaptive neural network controller is specifically:

[0018]

[0019] Wherein, k 2i >0,

[0020] In at least one embodiment, the experience-based learning controller is specifically:

[0021]

[0022] Wherein, [k a ,k b ] is a time interval after the system is stably converged.

[0023] In at least one embodiment, the experience-based learning controller is used to obtain knowledge of different users and different environmental conditions. The knowledge is fused by using a least square method based on spatial constraints, and a weight value after knowledge fusion is obtained:

[0024]

[0025] In the formula, A1 is a linearly independent component of A under subspace constraints,

[0026] In at least one embodiment, the learning controller based on knowledge fusion experience is constructed by using the weight value after knowledge fusion as:

[0027] In at least one embodiment, an admittance control is adopted, a motion trajectory is reshaped according to a dynamic relationship between an interaction torque and a trajectory deviation, a variation degree of stiffness and damping is determined, an external force is converted into a desired displacement, and the desired displacement is taken as a reference trajectory of the learning controller based on knowledge fusion experience.

[0028] In another aspect, the technical scheme of the present application also provides a discrete neural network control system of an upper limb rehabilitation robot, comprising:

[0029] The experience-based learning controller construction module is configured to: regard each joint module of the upper limb rehabilitation robot as a subsystem, construct an adaptive neural network controller based on a discrete barrier Lyapunov function, determine a weight update rate, and construct an experience-based learning controller according to a discrete deterministic learning theory.

[0030] The knowledge fusion experience-based learning controller construction module is configured to: on the basis of the experience-based learning controller, realize knowledge fusion by using a space constraint-based least square method, and construct a knowledge fusion experience-based learning controller.

[0031] The variable admittance control module is configured to: construct a variable admittance control model, take the output of the variable admittance control model as a reference trajectory of the knowledge fusion experience-based learning controller, and execute corresponding rehabilitation action training according to the rehabilitation needs of a user.

[0032] The technical scheme of the present application has the following advantages:

[0033] 1) The present application adopts a discrete barrier Lyapunov function to construct a control strategy, which can ensure that the system state is always constrained in a safe range, avoid secondary damage caused by the upper limb rehabilitation robot exceeding the movement range, and enhance the stability of the control system in a complex rehabilitation environment. On the other hand, the present application adopts a decentralized neural network control based on deterministic learning, which effectively reduces the number of neural network nodes required for centralized control, reduces the computational complexity, improves the real-time computing capability of the system, and retains the ability to accurately model the unknown dynamics in the system during the adaptive process. The learned experience knowledge can be stored in the form of a constant neural network for subsequent use in the same or similar control tasks.

[0034] 2) The present application introduces a knowledge fusion mechanism based on a space constraint-based least square method, which can fuse the knowledge obtained from multiple tasks by deterministic learning, expand the knowledge representation domain, and improve the generalization and robustness of the experience-based controller and the system.

[0035] 3) The present application constructs a variable admittance control strategy. The method can adaptively adjust the virtual impedance parameters (damping and stiffness) according to the force exerted by the patient and the motion state, realize personalized rehabilitation training, and improve the comfort and effect of training. The variable admittance control strategy is combined with the experience-based controller based on the barrier Lyapunov function and the deterministic learning knowledge fusion mechanism to realize the compliance between the man and the machine, while ensuring that the motion is always within the constraint range, thereby ensuring safety. In addition, high-performance control is realized without updating the control parameters, thereby saving system energy and time cost. BRIEF DESCRIPTION OF DRAWINGS

[0036] The drawings accompanying the specification of this application form a part thereof, serve to further provide a further understanding of the application, and together with the description of the exemplary embodiments of the application, serve to explain the application, and do not constitute an improper limitation on the application.

[0037] Figure 1 is a flowchart of the discrete neural network control method of the upper limb rehabilitation robot of the present application;

[0038] Figure 2 is a schematic diagram of the overall framework of the experience-based learning controller constructed based on the discrete barrier Lyapunov function and the discrete deterministic learning in the discrete neural network control method of the upper limb rehabilitation robot of the present application;

[0039] Figure 3 is an explanatory diagram of the variable stiffness model of the present application;

[0040] Figure 4 is a schematic diagram of the overall framework of the experience-based learning controller constructed based on the knowledge fusion experience by combining the discrete deterministic learning and the knowledge fusion strategy in the discrete neural network control method of the upper limb rehabilitation robot of the present application;

[0041] Figure 5 is a specific implementation flowchart of the discrete neural network control method of the upper limb rehabilitation robot of the present application. DETAILED DESCRIPTION

[0042] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the application. Unless otherwise indicated, all technical and scientific terms used in the present application have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0043] As introduced in the background, the purpose of the present application is to overcome the shortcomings of the prior art, and to provide a discrete neural network control method and system for an upper limb rehabilitation robot, which realizes intelligent and safe high-performance control of the upper limb rehabilitation robot for different individuals and different recovery periods.

[0044] Example 1

[0045] In one typical embodiment of the present application, as shown in Figure 1 The embodiment discloses a discrete neural network control method for an upper limb rehabilitation robot, and the method comprises the following steps:

[0046] Each joint module of the upper limb rehabilitation robot is regarded as a subsystem, an adaptive neural network controller is constructed based on a discrete barrier Lyapunov function, a weight update rate is determined, and an experience-based learning controller is constructed according to a discrete deterministic learning theory.

[0047] Based on the experience-based learning controller, knowledge fusion is realized by using a spatial constraint-based least square method, and an experience-based learning controller based on knowledge fusion is constructed.

[0048] A variable-inductance control model is constructed, the output of the variable-inductance control model is taken as a reference trajectory of the experience-based learning controller based on knowledge fusion, and corresponding rehabilitation action training is performed according to the rehabilitation requirements of a user.

[0049] The discrete neural network control method for the upper limb rehabilitation robot is described in detail.

[0050] In step S1, each joint module of the upper limb rehabilitation robot is regarded as a subsystem, an adaptive neural network controller is constructed based on a discrete barrier Lyapunov function, a weight update rate is determined, and an experience-based learning controller is constructed according to a discrete deterministic learning theory.

[0051] The existing dynamic model of the upper limb rehabilitation robot is constructed by taking the whole robot as a whole system. When a neural network is constructed for control, with the increase of the degrees of freedom, the dimension of the input of the neural network also increases, and the calculation burden also increases exponentially. In order to reduce the calculation burden, the embodiment first constructs a discrete dynamic model of the upper limb rehabilitation robot.

[0052] Specifically, the discrete-time dynamic model of the n-degree-of-freedom upper limb rehabilitation robot is considered as follows:

[0053]

[0054] Wherein, k is a discrete time point, T is a discrete time interval; q k ,v k represent the joint position and the joint speed, C(q k ,v k ) is a Coriolis-centrifugal force matrix, M(q k ) and is an inertia matrix; G(q k ) is a gravity torque matrix; τ k is a control input torque.

[0055] As Figure 2 shown, in order to reduce the computational burden, each standard joint module in (1) is regarded as a subsystem, which is connected by the coupling torque of the whole robot system. By separating the independent local variable from the coupling between the subsystems, the dynamics of the subsystem can be re-expressed in the joint space as:

[0056]

[0057] where, represents the coupling torque, which is specifically expressed as:

[0058]

[0059] where, and are the i-th element of q k , v k , and τ k vectors, respectively. and A ij (q k , v k ) are the i j-th element of A and A(q k , v k ) matrices, respectively.

[0060] Further, in order to avoid the non-causal problem that may occur in robot control, the coupling torque (3) is transformed into an output feedback system, which is converted into a dynamic model with the form of a standard system with output feedback, which is specifically expressed as:

[0061]

[0062] where, and

[0063]

[0064] After building the decentralized dynamics model of the upper limb rehabilitation robot, considering the symmetric potential barrier Lyapunov function, a virtual controller and an ideal adaptive neural network controller are constructed.

[0065] Specifically, the tracking errors are defined as and and are the joint position and velocity tracking errors, is the joint position reference trajectory; is the virtual controller.

[0066] Consider the following symmetric potential barrier Lyapunov function:

[0067]

[0068] where k bi is the i-th joint constraint boundary, is the estimation error of neural network weights. By taking the difference of the symmetric potential barrier Lyapunov function (i.e. equation (5)), the forms of the virtual controller and the ideal adaptive neural network controller are obtained as follows:

[0069]

[0070] where k 2i > 0, neural networks are used to approximate the unknown dynamics of each joint.

[0071] The weight update law is chosen as follows:

[0072]

[0073] where Γ i and σ i are the design parameters of the controller and are both positive integers.

[0074] By choosing appropriate design parameters, the system stability can be ensured, and the relevant state vectors converge to the following region:

[0075]

[0076] According to the certainty learning theory, for any periodic or recurrent trajectory, the regression sub-vector composed of the radial basis function neural network (RBFNN) almost always satisfies the persistent excitation condition. Based on the satisfaction of the persistent excitation condition, the local accurate neural network identification of the nonlinear system dynamics along the recurrent trajectory is realized, and the neural network weights will converge to their true values in the adaptive process. The learned knowledge is expressed in a time-invariant and spatially distributed manner and stored in the form of a constant RBFNN. The learned knowledge is specifically expressed as:

[0077]

[0078] where [k a , k b ] is the time interval after the system stable convergence.

[0079] The experience-based learning controller is constructed based on the learned knowledge as follows:

[0080]

[0081] The stored knowledge is an accurate neural network modeling of the nonlinear system dynamics, and in the subsequent same or similar tasks, the unknown dynamics of the system can be regarded as known information, and then the experience-based learning controller (11) designed based on this can obtain more excellent control performance.

[0082] However, it is worth noting that the knowledge obtained at present can only cover a narrow area near the actual motion trajectory of the system in the state space, and cannot form knowledge representation of a larger area. When there is interaction between the patient and the rehabilitation robot, the tracking trajectory or the system state changes, and the control performance will be poor.

[0083] Therefore, the embodiment constructs a knowledge fusion experience-based learning controller, that is, step S2 is performed.

[0084] Step S2, on the basis of the experience-based learning controller, the knowledge is fused by using the space constraint-based least square method to construct the knowledge fusion experience-based learning controller.

[0085] Firstly, the ideal adaptive neural network controller (7) and the learned knowledge (10) are used to obtain knowledge under different users and different environmental conditions, and the knowledge satisfies the following equation:

[0086]

[0087] Wherein, εi is the approximation error, W F is the fused weight.

[0088] Then, the knowledge is fused by using the space constraint-based least square method, and the fused weight is:

[0089]

[0090] In the formula, A1 is the linearly independent component of A under the subspace constraint, The space constraint-based least square method not only has low computational complexity, but also can maintain the local characteristics of the RBFNN in the control process.

[0091] Further, the knowledge fusion experience-based learning controller is constructed by using the fused weight:

[0092]

[0093] Because the fused weight W FThe dynamic information under different operating conditions is integrated to realize wide-area coverage of the dynamic model, so the learning controller (14) based on knowledge fusion experience has good generalization ability. For the control task of unknown dynamics in the extended knowledge area, the experience-based learning controller has better transient performance and higher control accuracy, and since the online calculation of the controller parameters is no longer needed, the system energy and control time are saved, and the experience-based learning controller also retains the constraint performance of barrier Lyapunov function. This has extremely important value for the active rehabilitation training of patients.

[0094] Step S3, a variable admittance control model is constructed, the output of the variable admittance control model is taken as the reference trajectory of the learning controller based on knowledge fusion experience, and the training of the corresponding rehabilitation action according to the rehabilitation needs of the user is performed.

[0095] In the process of assisting the patient in movement by the rehabilitation robot, the robot needs to track the task trajectory and realize compliance in physical human-robot interaction. Admittance control is a commonly used control method in the field of rehabilitation robots, but it is difficult for the admittance control with fixed parameters to balance safety and training effect. The variable admittance control adjusts the admittance parameters (such as virtual mass, damping and stiffness) online, so that the robot can adapt to the change of the patient's strength in different rehabilitation stages, and realize more natural human-robot interaction.

[0096] In order to ensure the safety of interaction, admittance control is adopted, and the dynamic relationship between the interaction torque F and the trajectory deviation e r = q r -q d The remodeled movement trajectory is specifically:

[0097]

[0098] Wherein, q d represents the expected trajectory, q r is the robot reference trajectory, which is taken as the input of the inner position control loop. M, B, K and F represent the inertia matrix, damping matrix, stiffness matrix and contact force of the second-order impedance model respectively. Among them, B and K are real-time variable.

[0099] Further, the change model of the stiffness is as follows:

[0100] K = (1 - a)K f + K s (16)

[0101] Where a = tanh(bF T F), b is a constant determining the change rate of a, K f and K s are the upper and lower limits of K respectively. Figure 3The stiffness variation pattern based on the external torque is shown.

[0102] Further, the variable damping design is as follows:

[0103]

[0104] In the formula, ξ is an adjustment gain. Through the above design, the robot can adapt to the rehabilitation needs of different patients by adjusting the damping and stiffness. For example, in upper limb rehabilitation, when the patient's strength is weak, the robot can reduce the stiffness and damping to provide assisted movement; and when the patient gradually recovers strength, the robot can increase the stiffness and damping to guide it to actively complete the task.

[0105] Then, the external force is converted into the expected displacement by the variable compliance model, which is then taken as the reference trajectory of the knowledge fusion experience-based learning controller to perform corresponding rehabilitation action training according to the user's rehabilitation needs. Based on the above operation, the rehabilitation robot can not only adapt to the external interaction force and ensure the movement within the constraint range, but also achieve high-performance control without updating the control parameters, and the specific process is as shown in Figure 4 . Figure 5 The overall process of the method and the result schematic diagram of an example are shown.

[0106] Embodiment 2

[0107] In a typical embodiment of the present application, the embodiment discloses a discrete neural network control system for an upper limb rehabilitation robot, comprising:

[0108] The experience-based learning controller construction module is configured to: regard each joint module of the upper limb rehabilitation robot as a subsystem, construct an adaptive neural network controller based on a discrete barrier Lyapunov function; determine a weight update rate, and construct an experience-based learning controller according to the discrete determination learning theory;

[0109] The knowledge fusion experience-based learning controller construction module is configured to: on the basis of the experience-based learning controller, realize the fusion of knowledge by using a spatial constraint-based least square method, and construct a knowledge fusion experience-based learning controller;

[0110] The variable compliance control module is configured to: construct a variable compliance control model, take the output of the variable compliance control model as the reference trajectory of the knowledge fusion experience-based learning controller, and perform corresponding rehabilitation action training according to the user's rehabilitation needs.

[0111] Embodiment 3

[0112] In a typical embodiment of the present application, the embodiment provides a computer readable storage medium, which stores a computer program, the program being executed by a processor to implement the steps in a discrete neural network control method for an upper limb rehabilitation robot as introduced in Embodiment 1, the steps comprising:

[0113] Each joint module of the upper limb rehabilitation robot is regarded as a subsystem, an adaptive neural network controller is constructed based on a discrete barrier Lyapunov function, a weight update rate is determined, and an experience-based learning controller is constructed according to a discrete deterministic learning theory;

[0114] Based on the experience-based learning controller, knowledge fusion is realized by using a spatial constraint-based least square method, and an experience-based learning controller based on knowledge fusion is constructed;

[0115] A variable admittance control model is constructed, the output of the variable admittance control model is taken as a reference trajectory of the experience-based learning controller based on knowledge fusion, and a corresponding rehabilitation action training is performed according to a rehabilitation requirement of a user.

[0116] Embodiment 4

[0117] In a typical embodiment of the present application, the embodiment provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps in a discrete neural network control method for an upper limb rehabilitation robot as introduced in Embodiment 1 when executing the program, the steps comprising:

[0118] Each joint module of the upper limb rehabilitation robot is regarded as a subsystem, an adaptive neural network controller is constructed based on a discrete barrier Lyapunov function, a weight update rate is determined, and an experience-based learning controller is constructed according to a discrete deterministic learning theory;

[0119] Based on the experience-based learning controller, knowledge fusion is realized by using a spatial constraint-based least square method, and an experience-based learning controller based on knowledge fusion is constructed;

[0120] A variable admittance control model is constructed, the output of the variable admittance control model is taken as a reference trajectory of the experience-based learning controller based on knowledge fusion, and a corresponding rehabilitation action training is performed according to a rehabilitation requirement of a user.

[0121] The above merely describes preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A discrete neural network control method for an upper limb rehabilitation robot, characterized in that, include: Each joint module of the upper limb rehabilitation robot is regarded as a subsystem, and an adaptive neural network controller is constructed based on the discrete obstacle Lyapunov function. Determine the weight update rate, construct an experience-based learning controller based on discrete learning theory; Based on the experience-based learning controller, knowledge fusion is achieved by using the least squares method based on spatial constraints, thus constructing a learning controller based on knowledge fusion experience. A variable admittance control model is constructed, and the output of the variable admittance control model is used as the reference trajectory of a learning controller based on knowledge fusion experience. The corresponding rehabilitation actions are trained according to the user's rehabilitation needs. Utilize an experience-based learning controller to acquire knowledge from different users and under different environmental conditions. Knowledge fusion is achieved using the least squares method based on spatial constraints, resulting in the weights after knowledge fusion: In the formula , A 1 is under subspace constraints A The linear independent components, .

2. The discrete neural network control method for an upper limb rehabilitation robot as described in claim 1, characterized in that, Each standard joint module in the upper limb rehabilitation robot is considered as a subsystem and interconnected through the coupling torque of the entire upper limb rehabilitation robot system; the coupling torque is transformed into an output feedback system, which is converted into a dynamic model in the form of a canonical system with output feedback; considering the symmetric barrier Lyapunov function, a virtual controller and an ideal adaptive neural network controller are constructed.

3. The discrete neural network control method for an upper limb rehabilitation robot as described in claim 2, characterized in that, The ideal adaptive neural network controller is specifically: in, , .

4. The discrete neural network control method for an upper limb rehabilitation robot as described in claim 3, characterized in that, The experience-based learning controller is specifically as follows: in, , This refers to the time interval after the system has reached stable convergence.

5. The discrete neural network control method for an upper limb rehabilitation robot as described in claim 1, characterized in that, Using the weights obtained after knowledge fusion, a learning controller based on knowledge fusion experience is constructed as follows: 。 6. The discrete neural network control method for an upper limb rehabilitation robot as described in claim 1, characterized in that, Admittance control is employed to reshape the motion trajectory based on the dynamic relationship between the interaction torque and the trajectory deviation, determine the degree of change in stiffness and damping, and convert the external force into the desired displacement, which is then used as the reference trajectory for a learning controller based on knowledge fusion experience.

7. A discrete neural network control system for an upper limb rehabilitation robot, characterized in that, include: The experience-based learning controller building module is configured to treat each joint module of the upper limb rehabilitation robot as a subsystem and build an adaptive neural network controller based on the discrete obstacle Lyapunov function. Determine the weight update rate, construct an experience-based learning controller based on discrete learning theory; The module for building a learning controller based on knowledge fusion experience is configured to: build a learning controller based on knowledge fusion experience by using the least squares method based on spatial constraints to achieve knowledge fusion on the basis of the experience-based learning controller. The variable admittance control module is configured to: construct a variable admittance control model, use the output of the variable admittance control model as the reference trajectory of the learning controller based on knowledge fusion experience, and execute corresponding rehabilitation action training according to the user's rehabilitation needs; Utilize an experience-based learning controller to acquire knowledge from different users and under different environmental conditions. Knowledge fusion is achieved using the least squares method based on spatial constraints, resulting in the weights after knowledge fusion: In the formula , A 1 is under subspace constraints A The linear independent components, .

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the discrete neural network control method for an upper limb rehabilitation robot as described in any one of claims 1-6.

9. A computer device, 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 implements the steps in the discrete neural network control method for an upper limb rehabilitation robot as described in any one of claims 1-6.

Images