Control device, control method for rehabilitation robot, and rope-pulling rehabilitation robot

Abstract

The application relates to the technical field of rehabilitation robot human-computer interaction and intelligent control, and discloses a control device and a control method of a rehabilitation robot and a rope traction rehabilitation robot, which comprise: a high-layer error distribution remodeling model, which is used for calculating an expected error distribution gradient and ergodicity measurement of a human upper limb motion trajectory, and calculating a virtual auxiliary force according to the expected error distribution gradient and the ergodicity measurement; and a low-layer admittance control model, which is used for obtaining a target motion trajectory according to an interactive force signal, the virtual auxiliary force and the human upper limb motion trajectory, and converting the target motion trajectory into motor control quantity of the rehabilitation robot, so as to control motor output of the rehabilitation robot. The application evaluates the motion performance of a subject by using the ergodicity measurement, and guides the motion decision by taking the ergodicity measurement reduction as a target, so that the ergodicity measurement and the trajectory error index are reduced, the ergodicity and the motion performance of the subject in the interactive motion are improved, and the motion function of a patient in rehabilitation training can be recovered.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction and intelligent control technology for rehabilitation robots, and in particular to control devices, control methods, and rope-traction rehabilitation robots for rehabilitation robots. Background Technology

[0002] Currently, the most common human-machine interaction control methods for upper limb rehabilitation robots include traditional control methods such as admittance control, impedance control, and virtual path control. While these traditional control methods can adjust robot assistance based on positional errors and the dynamic relationship between human-machine interaction forces during rehabilitation training, they do not fully respect the characteristics of natural human-machine interaction, ignore the natural variability of human movement, and are not conducive to the subject's active movement exploration, thus affecting the recovery of motor function.

[0003] First, these traditional control methods typically employ fixed, pre-defined motion tasks along predetermined trajectories, restricting human-computer interaction to a fixed time-series motion trajectory and limiting the subject's freedom of movement. Second, traditional control methods often prioritize reducing trajectory tracking errors, neglecting the natural variability of movement during rehabilitation training, which may limit further improvements in robot-assisted rehabilitation training effectiveness. Finally, in complex human-computer interaction environments, using motion trajectory error to assess motion performance has low resolution, making it difficult to distinguish changes in motion performance caused by the subject's motor ability deficit or assistance. Furthermore, the motion performance assessment information is susceptible to random interference, thus affecting the optimization of real-time control parameters.

[0004] For the reasons mentioned above, current traditional control methods limit the subjects' ability to actively explore movement, which may limit further improvement in the effectiveness of robot-assisted rehabilitation training. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a control device, control method, and rope-traction rehabilitation robot for reshaping error distribution based on ergodicity measurement. The ergodicity measurement is used to assess the subject's motor performance, and the reduction of ergodicity measurement is used to guide motor decisions, thereby improving the subject's ergodicity and motor performance in interactive movements and helping to restore the patient's motor function during rehabilitation training.

[0006] In a first aspect, the present invention provides a control device for a rehabilitation robot, the device comprising:

[0007] Interconnected information acquisition systems, high-level error distribution reshaping models, and low-level admittance control models;

[0008] The information acquisition system is used to collect the human upper limb movement trajectory and the interaction force signal applied by the human upper limb to the rehabilitation robot in real time during human-computer interaction, send the human upper limb movement trajectory to the high-level error distribution reshaping model, and send the human upper limb movement trajectory and the interaction force signal to the low-level admittance control model.

[0009] The high-level error distribution reshaping model is used to calculate the expected error distribution gradient and ergodicity measure of the human upper limb movement trajectory, calculate the virtual assist force based on the expected error distribution gradient and the ergodicity measure, and send the virtual assist force to the low-level admittance control model.

[0010] The low-level admittance control model is used to obtain the target motion trajectory based on the interaction force signal, the virtual assist force, and the human upper limb motion trajectory, and convert the target motion trajectory into the motor control quantity of the rehabilitation robot to control the motor output of the rehabilitation robot.

[0011] Furthermore, the high-level error distribution reshaping model includes interconnected expected error distribution models, error distribution gradient models, and ergodic estimation motion performance models;

[0012] The expected error distribution model is used to statistically analyze the movement trajectory of the human upper limb using a Gaussian mixture model to obtain the expected error distribution probability density, and then sends the expected error distribution probability density to the expected error distribution model and the ergodic estimation motion performance model respectively.

[0013] The expected error distribution model is used to calculate the gradient of the expected error distribution probability density, thereby obtaining the expected error distribution gradient.

[0014] The ergodic estimation motion performance model is used to calculate the distance between the spatial Fourier coefficients of the human upper limb movement trajectory and the Fourier coefficients of the expected error distribution probability density, and the ergodicity metric is calculated using the spectral method.

[0015] Furthermore, the high-level error distribution reshaping model is also used to obtain the virtual auxiliary force based on the product of the expected error distribution gradient and the ergodicity metric, wherein the magnitude of the virtual auxiliary force is determined by the ergodicity metric, and the direction of the virtual auxiliary force is determined by the expected error distribution gradient.

[0016] Furthermore, the probability density of the expected error distribution is calculated using the following formula:

[0017]

[0018] In the formula, k is the dimension, x iLet be the error state variable of the i-th dimension, s be the covariance matrix, μ be the mean, and T be the total motion time;

[0019] The ergodicity metric is calculated using the following formula:

[0020]

[0021] In the formula, Let c be the Fourier coefficient of the expected error distribution probability density. k Here, K represents the time-averaged coefficients of the Fourier coefficients of the basis functions for the trajectory of the human upper limbs, and Λ represents the number of basis functions in each dimension. k For weights.

[0022] Furthermore, the low-level admittance control model includes an admittance filter and a position controller connected to the admittance filter;

[0023] The admittance filter is used to perform admittance filtering on the interaction force signal and the virtual auxiliary force to obtain the desired motion trajectory, and then send the desired motion trajectory to the position controller.

[0024] The position controller is used to perform position control on the human upper limb movement trajectory and the desired movement trajectory to obtain the target movement trajectory, and convert the target movement trajectory as a control quantity into the motor control quantity of the rehabilitation robot to control the motor output of the rehabilitation robot.

[0025] Furthermore, the desired motion trajectory is represented by the following formula:

[0026]

[0027] In the formula, M d D is a zero matrix. d Let B be the stiffness matrix. d Let F be the damping matrix. h For the interaction force signal, F r P is a virtual auxiliary force. d The desired trajectory.

[0028] Furthermore, the position controller is a fast finite-time convergence controller.

[0029] Secondly, the present invention provides a control method for a rehabilitation robot, the method comprising:

[0030] Real-time acquisition of human upper limb movement trajectories and force signals exerted by human upper limbs on rehabilitation robots during human-computer interaction;

[0031] Calculate the expected error distribution gradient and ergodicity measure of the human upper limb movement trajectory, and calculate the virtual assist force based on the expected error distribution gradient and the ergodicity measure;

[0032] Based on the interactive force signal, the virtual assistive force, and the human upper limb movement trajectory, the target movement trajectory is obtained, and the target movement trajectory is converted into the motor control quantity of the rehabilitation robot to control the motor output of the rehabilitation robot.

[0033] Furthermore, the steps of calculating the expected error distribution gradient and ergodicity measure of the human upper limb movement trajectory include:

[0034] The expected error distribution probability density was obtained by statistically analyzing the human upper limb movement trajectory using a Gaussian mixture model.

[0035] Calculate the gradient of the probability density of the expected error distribution to obtain the gradient of the expected error distribution;

[0036] The distance between the Fourier coefficients of the human upper limb movement trajectory and the Fourier coefficients of the expected error distribution probability density is calculated, and the ergodicity metric is obtained by using the spectral method.

[0037] Thirdly, embodiments of the present invention also provide a rope traction rehabilitation robot, which stores a computer program that, when executed by a processor, implements the steps of the above-described method.

[0038] This invention provides a rehabilitation robot control device, control method, and rope traction rehabilitation robot. Taking into account the natural variability of the subject's movement, this invention uses ergodicity measurement combined with an error distribution gradient model to customize a virtual assistive force. Under the action of this force, the movement trajectory can be transferred from a low-probability-density space to a high-density space according to the spatial distribution characteristics of the desired error distribution, thus reshaping the error distribution. This can improve the subject's ergodicity and motor performance in interactive movements, and help restore the patient's motor function during rehabilitation training. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the control device of the rehabilitation robot in an embodiment of the present invention;

[0040] Figure 2 This is a flowchart illustrating the control method of the rehabilitation robot in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the apparatus used in the comparative experiment in an embodiment of the present invention;

[0042] Figure 4 This is a diagram of the expected error distribution model constructed based on motion trajectory data in the comparative experiment;

[0043] Figure 5 This is a graph showing the error distribution results of the subjects in the comparative experiment;

[0044] Figure 6 This is a graph showing the mean and variance of the ergodic measures of the subjects in a comparative experiment. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 The first embodiment of this invention proposes a control device for a rehabilitation robot. This invention is mainly used to control an upper limb rope traction rehabilitation robot, comprising an interconnected information acquisition system 1, a high-level error distribution reshaping model 2, and a low-level admittance control model 3. The information acquisition system 1 is used to collect in real time the human upper limb movement trajectory and the interaction force signal applied by the human upper limb to the rehabilitation robot during human-computer interaction. It sends the human upper limb movement trajectory to the high-level error distribution reshaping model 2 and the human upper limb movement trajectory and interaction force signal to the low-level admittance control model 3. The high-level error distribution reshaping model 2 is used to calculate the expected error distribution gradient and ergodicity measure of the human upper limb movement trajectory, calculate the virtual assist force based on the expected error distribution gradient and ergodicity measure, and send the virtual assist force to the low-level admittance control model 3. The low-level admittance control model 3 is used to obtain the target movement trajectory based on the interaction force signal, the virtual assist force, and the human upper limb movement trajectory, and convert the target movement trajectory into a motor control quantity for the rehabilitation robot to control the motor output of the rehabilitation robot. The various parts of this device are described in detail below.

[0047] This invention collects human upper limb movement data through an information acquisition system 1. Specifically, it can reset the end effector of the upper limb rope traction rehabilitation robot to a designated starting position, guide the user to place the upper limb on the end effector of the rope traction rehabilitation robot, guide the user to become familiar with rehabilitation training movements through an interactive interface, and then collect relevant data of the user's movement trajectory in real time. In addition, it collects the interaction force signal of the user's upper limb applied to the rope traction rehabilitation robot in real time through a force sensor installed on the rope, and sends the collected movement trajectory to the high-level error distribution reshaping model 2, and sends the movement trajectory and interaction force signal to the low-level admittance control model 3.

[0048] The high-level error distribution reshaping model 2 takes real-time kinematic feedback information as input and outputs virtual auxiliary force. In this embodiment, the high-level error distribution reshaping model 2 includes an interconnected expected error distribution model 21, an error distribution gradient model 22, and an ergodic estimation motion performance model 23.

[0049] Specifically, the expected error distribution model 21 statistically analyzes the acquired human upper limb movement trajectory to obtain the expected error distribution probability density. Preferably, a Gaussian mixture model can be used to construct the expected error distribution probability density function, that is, the histogram2 function in MATLAB is applied to statistically analyze the relevant data of the movement trajectory to obtain the expected error distribution probability density function. The expression obtained by using MATLAB's gmdistribution function for maximum likelihood estimation is as follows:

[0050]

[0051] In the formula, k is the dimension, x i Let be the error state variable of the i-th dimension, s be the covariance matrix, μ be the mean, and T be the total motion time.

[0052] The expected error distribution probability density obtained in the expected error distribution model 21 is sent to the error distribution gradient model 22 and the ergodic estimation motion performance model 23, respectively. The error distribution gradient model 22, upon receiving the expected error distribution probability density function... Then, its gradient is calculated to obtain the gradient of the expected error distribution.

[0053]

[0054] Where, e = [x1,…,x k The expected error distribution gradient is used to generate a direction vector from a low probability density spatial distribution to a high probability density spatial distribution.

[0055] In the ergodic estimation model 23, the spectral method is used to define the ergodicity measure. Specifically, the similarity between the spatial Fourier coefficients of the human upper limb movement trajectory and the Fourier coefficients of the probability density function of the expected error distribution is measured by comparing the distance between the two distributions. The ergodic measure is used to quantify the amount of information encoded in the real-time movement trajectory for the movement task, and the performance of the subject's movement is evaluated based on the overall target feature information. The calculation formula is as follows:

[0056]

[0057] In the formula, Let c be the Fourier coefficient of the expected error distribution probability density. kHere, K represents the time-averaged coefficients of the Fourier coefficients of the basis functions for the trajectory of the human upper limbs, and Λ represents the number of basis functions in each dimension. k For the weights, where,

[0058] Λ k =(1+|k|) 2 ) -s

[0059]

[0060] The Fourier coefficients of the expected error distribution probability density are calculated using the inner product. It can be represented as:

[0061]

[0062] The time-averaged coefficients of the Fourier coefficients of the basis functions of the human upper limb movement trajectory can be expressed as:

[0063]

[0064] The Fourier basis functions are expressed in the following form:

[0065]

[0066] In the above formula, X is the range of parameter x, x(t) is the trajectory of the human upper limb movement, and h k L is the normalization factor, n is the number of dimensions, and L is the normalization factor. i Let Λ be the length of the i-th dimension state. k It allows for greater weighting of low-frequency information.

[0067] After obtaining the expected error distribution gradient and ergodicity measure, the high-level error distribution reshaping model 2 can calculate the virtual assistive force by multiplying the two. The magnitude of the virtual assistive force is determined by the ergodicity measure, while its direction is determined by the expected error distribution gradient. This invention considers the natural variability of subject movement and uses an ergodicity measure combined with an error distribution gradient model to customize the virtual assistive force. Under the action of this virtual assistive force, the movement trajectory can be transferred from a low-probability-density space to a high-density space according to the spatial distribution characteristics of the expected error distribution, thus reshaping the error distribution. This can improve the ergodicity and motor performance of subjects in interactive movements, and help restore patients' motor function in rehabilitation training.

[0068] In this embodiment of the invention, the low-level admittance control model 3 takes the virtual auxiliary force generated by the high-level error distribution reshaping model 2 and the interactive force signal collected by the information acquisition system 1 as inputs. After data processing, the output data is the motor control quantity of the rehabilitation robot. The low-level admittance control model 3 includes an admittance filter 31 and a position controller 32 connected to it.

[0069] The admittance filter 31 is used to perform admittance filtering on the input interaction force signal and the virtual auxiliary force. Its output is the desired motion trajectory, which can be expressed by the following formula:

[0070]

[0071] In the formula, M d D is a zero matrix. d Let B be the stiffness matrix. d Let F be the damping matrix. h For the interaction force signal, F r P is a virtual auxiliary force. d The desired trajectory.

[0072] The desired motion trajectory generated by the admittance filter 31 is input to the position controller 32. The position controller 32 determines the target motion trajectory according to the input desired motion trajectory and the motion trajectory of the human upper limb. Then, the determined target motion trajectory is used as the control quantity. Combined with the dynamic model of the rehabilitation robot, the corresponding motor control quantity is generated to control the motor output of the rehabilitation robot. Preferably, a fast finite-time convergence controller can be used to convert the position control and motor control quantities. The specific data processing process can refer to the conventional position controller processing method, which will not be elaborated here.

[0073] Please see Figure 2 Based on the same inventive concept, the second embodiment of the present invention proposes a rehabilitation robot control method, comprising:

[0074] Step S10: Real-time acquisition of the human upper limb movement trajectory and the interaction force signal applied by the human upper limb to the rehabilitation robot during human-computer interaction;

[0075] Step S20: Calculate the expected error distribution gradient and ergodicity measure of the human upper limb movement trajectory, and calculate the virtual assist force based on the expected error distribution gradient and the ergodicity measure;

[0076] Step S30: Based on the interactive force signal, the virtual assistive force, and the human upper limb movement trajectory, the target movement trajectory is obtained, and the target movement trajectory is converted into the motor control quantity of the rehabilitation robot to control the motor output of the rehabilitation robot.

[0077] Further, step S20 includes:

[0078] Step S201: The trajectory of the human upper limb movement is statistically analyzed using a Gaussian mixture model to obtain the expected error distribution probability density;

[0079] Step S202: Calculate the gradient of the probability density of the expected error distribution to obtain the gradient of the expected error distribution;

[0080] Step S203: Calculate the distance between the Fourier coefficients of the human upper limb movement trajectory and the Fourier coefficients of the expected error distribution probability density, and use the spectral method to calculate the ergodicity measure.

[0081] This invention provides a rehabilitation robot control method that extracts motion features from upper limb motion data of healthy individuals and encodes common, global, and naturally variable information related to motion into the target motion task using statistical distributions such as error distributions. This approach breaks away from the strict constraint of training tasks on fixed time-series position or torque signals, making the training motion task universal and similar to common characteristics of human movement. This guides rehabilitation training and promotes natural and coordinated human-robot interaction. Furthermore, it employs ergodic metrics to evaluate real-time interactive motion performance and adjusts the control method to increase the spatial statistics in the interactive motion that align with the desired error distribution, thereby promoting subject motor learning. The control method provided by this invention generates a reference distribution through a series of non-standard motion demonstrations, avoiding consideration of highly standardized trajectories or inferences about which motion parameters are task-related, enabling the human-robot coupled system to complete tasks under different constraints.

[0082] The technical effectiveness of this invention is verified by comparing two control methods: error distribution reshaping control based on ergodicity metric provided by this invention and conventional virtual path control.

[0083] First, 10 healthy subjects were recruited to participate in a human-computer interaction movement experiment assisted by an upper limb rope traction rehabilitation robot. Movement trajectory data were collected to construct the expected error distribution. like Figure 3 As shown, the upper limb traction rehabilitation robot provides visual feedback. The position of the robot's end effector is mapped onto the screen as a cursor, with a pre-defined circular reference path. Subjects control the robot's end effector, using the cursor as a pen to trace the reference path on the XZ plane. Each subject is required to make their best effort to move the cursor accurately and quickly counterclockwise along the reference path, with a task time of 200 seconds. The experiment collected 2,000,000 data points from 10 subjects, and this data was used to construct an error distribution model. Specifically, the histogram2 function in MATLAB R2020a was used to calculate the error distribution model, as shown in Figure 4.

[0084] Twelve healthy subjects were then recruited, and a circular drawing experiment under random disturbances was conducted using two methods: the error distribution reshaping control based on ergodicity metric provided in this invention and the conventional virtual path control method. The output virtual assist force Fr was generated using both methods and input into the low-level admittance control model to control the rehabilitation robot. The performance of the two control methods was compared according to the following evaluation metrics:

[0085] (1) The root mean square error (RMSD) of the trajectory is defined as:

[0086]

[0087] (2) Ergodicity measure ε, the smaller the index, the more similar the error distribution is to the expected distribution, and the higher the motion ergodicity:

[0088]

[0089] Figure 5 The error distribution of 12 subjects under the two methods is given by Figure 5 It can be seen that, compared with the virtual path control method, the error distribution reshaping control method based on ergodicity metric has a higher peak at zero point, higher density, and more clustered distribution.

[0090] Figure 6 The left and right plots show the mean and variance of the ergodic measure for 12 subjects under two different methods. The left plot represents the experimental results under the error distribution reshaping control method based on the ergodic measure, while the right plot represents the experimental results under the virtual path control method. The shaded areas represent the respective standard deviation bands. Figure 6 It can be seen that, compared to virtual path control, the error distribution reshaping control method based on ergodicity metrics shows that, as exercise training progresses, the ergodicity metric continuously decreases until it reaches a relatively low level after 50 seconds, and then remains at a relatively low level. While the virtual path control method also shows a slight initial decrease, the ergodicity metric remains at a higher level than the error distribution reshaping control method based on ergodicity metrics, and the fluctuation range is also larger. Therefore, compared to virtual path control, the control method provided by this invention can reduce ergodicity metrics and trajectory error indicators, indicating that this invention can improve the subject's motor ergodicity and motor performance, and helps to restore the patient's motor function in rehabilitation training.

[0091] Furthermore, this invention also proposes a rope traction rehabilitation robot, which stores a computer program that, when executed by a processor, implements the steps of the above-described method.

[0092] In summary, the present invention provides a control device, control method, and rehabilitation robot for a rehabilitation robot. The device includes an interconnected information acquisition system, a high-level error distribution reshaping model, and a low-level admittance control model. The information acquisition system is used to acquire the motion trajectory of the human upper limb and the interaction force signal applied by the human upper limb to the rehabilitation robot, send the human upper limb motion trajectory to the high-level error distribution reshaping model, and send the human upper limb motion trajectory and the interaction force signal to the low-level admittance control model. The high-level error distribution reshaping model is used to calculate the expected error distribution gradient and ergodicity measure of the human upper limb motion trajectory, calculate a virtual assisting force based on the expected error distribution gradient and the ergodicity measure, and send the virtual assisting force to the low-level admittance control model. The low-level admittance control model is used to obtain a target motion trajectory based on the interaction force signal, the virtual assisting force, and the human upper limb motion trajectory, and convert the target motion trajectory into a motor control quantity for the rehabilitation robot to control the motor output of the rehabilitation robot. This invention assesses subjects’ motor performance using ergodicity metrics and guides motor decisions with the goal of reducing ergodicity metrics. It reduces ergodicity metrics and trajectory error indicators, improves subjects’ ergodicity and motor performance in interactive movements, and helps restore patients’ motor function in rehabilitation training.

[0093] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the method embodiments are basically similar to the device embodiments, so the description is relatively simple; refer to the description of the method embodiments for relevant details. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0094] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A control device for a rehabilitation robot, characterized in that, include: Interconnected information acquisition systems, high-level error distribution reshaping models, and low-level admittance control models; The information acquisition system is used to collect the human upper limb movement trajectory and the interaction force signal applied by the human upper limb to the rehabilitation robot in real time during human-computer interaction, send the human upper limb movement trajectory to the high-level error distribution reshaping model, and send the human upper limb movement trajectory and the interaction force signal to the low-level admittance control model. The high-level error distribution reshaping model is used to calculate the expected error distribution gradient and ergodicity measure of the human upper limb movement trajectory, calculate the virtual assist force based on the expected error distribution gradient and the ergodicity measure, and send the virtual assist force to the low-level admittance control model. The low-level admittance control model is used to obtain the target motion trajectory based on the interaction force signal, the virtual assist force, and the human upper limb motion trajectory, and convert the target motion trajectory into the motor control quantity of the rehabilitation robot to control the motor output of the rehabilitation robot. The high-level error distribution reshaping model includes interconnected expected error distribution models, error distribution gradient models, and ergodic estimation motion performance models. The expected error distribution model is used to statistically analyze the movement trajectory of the human upper limb using a Gaussian mixture model to obtain the expected error distribution probability density, and then sends the expected error distribution probability density to the expected error distribution model and the ergodic estimation motion performance model respectively. The expected error distribution model is used to calculate the gradient of the expected error distribution probability density, thereby obtaining the expected error distribution gradient. The ergodic estimation motion performance model is used to calculate the distance between the spatial Fourier coefficients of the human upper limb movement trajectory and the Fourier coefficients of the expected error distribution probability density, and the ergodicity metric is calculated using the spectral method. The high-level error distribution reshaping model is further used to obtain the virtual auxiliary force based on the product of the expected error distribution gradient and the ergodicity metric, wherein the magnitude of the virtual auxiliary force is determined by the ergodicity metric, and the direction of the virtual auxiliary force is determined by the expected error distribution gradient.

2. The control device for the rehabilitation robot according to claim 1, characterized in that, The expected error distribution probability density is calculated using the following formula: In the formula, k is the dimension. Let be the error state variable of the i-th dimension, s be the covariance matrix, μ be the mean, and T be the total motion time; The ergodicity metric is calculated using the following formula: In the formula, The Fourier coefficients are the expected error distribution probability density. K represents the time-averaged coefficients of the Fourier coefficients of the basis functions for the trajectory of human upper limb movement, where K is the number of basis functions in each dimension. For weights.

3. The control device for the rehabilitation robot according to claim 1, characterized in that, The low-level admittance control model includes an admittance filter and a position controller connected to the admittance filter; The admittance filter is used to perform admittance filtering on the interaction force signal and the virtual auxiliary force to obtain the desired motion trajectory, and then send the desired motion trajectory to the position controller. The position controller is used to perform position control on the human upper limb movement trajectory and the desired movement trajectory to obtain the target movement trajectory, and convert the target movement trajectory as a control quantity into the motor control quantity of the rehabilitation robot to control the motor output of the rehabilitation robot.

4. The control device for the rehabilitation robot according to claim 3, characterized in that, The desired motion trajectory is represented by the following formula: In the formula, It is a zero matrix. Here is the stiffness matrix. Here is the damping matrix. For interactive force signals, As a virtual auxiliary force, The desired trajectory.

5. The control device for the rehabilitation robot according to claim 4, characterized in that, The position controller is a fast finite-time convergence controller.

Images