A method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot

Abstract

The present invention discloses a method for learning and interactively controlling stability and compliance skills in a rehabilitation robot. The method comprises: obtaining rehabilitation therapy motion data during a rehabilitation therapy skill demonstration to construct a rehabilitation motion database; constructing a motion position-velocity skill model, a position-posture skill model, and a position-force / torque skill model for the rehabilitation robot based on the rehabilitation motion database to enable the rehabilitation robot to learn rehabilitation therapy skills; combining skill transfer and skill generalization to control the skills of the rehabilitation robot after learning the rehabilitation therapy skills; and implementing time-varying impedance control of the rehabilitation robot through the rehabilitation robot's motion trajectory error and real-time interactive force feedback during the rehabilitation therapy process. The present invention can achieve the generalization capability of the rehabilitation robot's rehabilitation therapy motion trajectory and meet the requirements for compliant control of the rehabilitation robot.

Description

Technical Field

[0001] The present invention relates to the field of rehabilitation robot motion control, and in particular to a method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot. Background Art

[0002] Motor dysfunction, such as hemiplegia, is common among stroke patients, creating an urgent need for rehabilitation-assisted treatment. Rehabilitation physiotherapy, an external treatment method based on Traditional Chinese Medicine (TCM) physiotherapy and guided by the theory of meridians, directly acts on localized areas of the body through physical factors, prompting interactions between nerves and body fluids to cause reactions in the human body, thereby helping to treat conditions such as stroke and sports injuries. However, traditional rehabilitation physiotherapy is heavily dependent on the physical labor of rehabilitation therapists, resulting in long treatment cycles and high costs. Furthermore, there is a severe shortage of rehabilitation therapists in my country, making it difficult to meet the urgent needs of patients for rehabilitation physiotherapy.

[0003] Rehabilitation robots can effectively reduce the workload of rehabilitation therapists, provide scientifically effective treatment plans, and help patients regain limb function, thereby improving their quality of life. Therefore, rehabilitation robots are one of the important ways to meet patients' urgent needs for rehabilitation therapy. There are many types of rehabilitation robots available, but most of them have limited rehabilitation methods and poor human-machine interaction, making it difficult to meet the rehabilitation therapy needs of different patients. Key shortcomings include: 1) Single movement method. Rehabilitation therapists pre-set movement trajectories and implement fixed movement methods, which are not generalizable. 2) Poor human-machine interaction. During the rehabilitation therapy process, it is impossible to adjust the rehabilitation therapy plan online based on patient feedback. 3) Poor compliance of rehabilitation robots. During the rehabilitation therapy process, it is impossible to adjust the stiffness of the rehabilitation robot online, making it difficult for rehabilitation robots to implement complex rehabilitation therapy environments.

[0004] Patent document CN117681202A discloses a method, system, and computer device for controlling the motion trajectory of a rehabilitation robot. This solution improves the accuracy of mathematical models containing model uncertainty by using a disturbance observer to approximate them. It also constrains tracking errors using a logarithmic barrier Lyapunov function, ensuring that the robotic arm moves within the desired trajectory range and avoiding secondary injuries to the patient. However, the rehabilitation robot using this method still suffers from problems such as a single motion mode and poor compliance. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot.

[0006] To achieve the above object, the technical solution of the present invention is:

[0007] A method for learning and interactively controlling stability and compliance skills of a rehabilitation robot, the method comprising:

[0008] Acquire rehabilitation physiotherapy movement data during the rehabilitation physiotherapy skill demonstration to build a rehabilitation movement database;

[0009] Constructing a motion position-velocity skill model, a position-posture skill model, and a position-force / torque skill model for the rehabilitation robot based on the rehabilitation motion database to enable the rehabilitation robot to learn rehabilitation therapy skills;

[0010] Combining skill transfer and skill generalization, the rehabilitation robot is regulated after learning rehabilitation therapy skills to prevent the rehabilitation robot from exceeding the treatment scope during the treatment task and to enable the rehabilitation robot to adapt to different rehabilitation therapy needs;

[0011] Time-varying impedance control of the rehabilitation robot is achieved through the motion trajectory error of the rehabilitation robot and the real-time interactive force feedback of the rehabilitation robot during rehabilitation therapy.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] The present invention constructs a comprehensive rehabilitation skill library for skill demonstration. Regarding skill learning, it uses skill learning methods to construct position-velocity skill models, position-posture skill models, and position-force / torque skill models for rehabilitation robots. Skill adjustment prevents the robot from exceeding the treatment range during treatment tasks, potentially harming the patient. Furthermore, to address the diverse rehabilitation needs of patients, the rehabilitation theory skills learned by the robot are generalized to a certain extent to better meet the needs of different patients. Finally, a time-varying impedance controller is designed to ensure compliant control of the rehabilitation robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of the method for learning and interactively controlling stability and compliance skills of a rehabilitation robot provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] Example:

[0016] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0017] See Figure 1 As shown, the method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot provided in this embodiment mainly includes the following steps:

[0018] Skill demonstration step: Acquire rehabilitation physiotherapy movement data during the rehabilitation physiotherapy skill demonstration process to build a rehabilitation movement database.

[0019] Skill learning step: constructing a motion position-speed skill model, a position-posture skill model and a position-force / torque skill model of the rehabilitation robot based on the rehabilitation motion database to realize the learning of rehabilitation physiotherapy skills of the rehabilitation robot.

[0020] In this way, through this step, the rehabilitation robot can accurately learn the motor skills of the rehabilitation therapist and realize the precise transmission of motor skills of "rehabilitation therapist-rehabilitation robot".

[0021] Skill adjustment steps: Combining skill transfer and skill generalization, the rehabilitation robot is regulated after learning rehabilitation therapy skills to prevent the rehabilitation robot from exceeding the treatment range during the treatment task and to enable the rehabilitation robot to adapt to different rehabilitation therapy needs.

[0022] In this way, through this step, the rehabilitation robot can be prevented from exceeding the treatment range during the treatment task process, so as to avoid the robot exceeding the treatment range during the treatment task process and causing harm to the patient; at the same time, in view of the differences in patients' rehabilitation needs, the rehabilitation theoretical skills learned and mastered by the rehabilitation robot are generalized to a certain extent to better meet the needs of different patients.

[0023] Variable impedance control steps: time-varying impedance control of the rehabilitation robot is achieved through the motion trajectory error of the rehabilitation robot and the real-time interactive force feedback of the rehabilitation robot during rehabilitation therapy.

[0024] Thus, a time-varying impedance controller is designed through this step to meet the compliant control of the rehabilitation robot.

[0025] As can be seen, the present invention constructs a comprehensive rehabilitation skill library for skill demonstration. Regarding skill learning, it constructs position-velocity skill models, position-posture skill models, and position-force / torque skill models for the rehabilitation robot through skill learning methods. Furthermore, skill adjustment prevents the rehabilitation robot from exceeding the treatment range during treatment tasks, thereby preventing harm to the patient. Furthermore, to address the diverse rehabilitation needs of patients, the rehabilitation theory skills learned by the rehabilitation robot are generalized to a certain extent to better meet the needs of different patients. Finally, a time-varying impedance controller is designed to ensure compliant control of the rehabilitation robot.

[0026] Specifically, in the above skill demonstration step, rehabilitation therapy exercise data is obtained in the following manner:

[0027] To meet the needs of rehabilitation therapy, a six-dimensional force sensor is installed on the robot's end to measure the interaction forces between the robot and its environment in real time. Using the robot's gravity compensation, a rehabilitation therapist uses their prior knowledge of rehabilitation therapy to interact with the robot and demonstrate tasks commonly used in rehabilitation therapy, including four main types of techniques: pressing, tapping, massage, and pushing. Among them, pressing: use the fingers or palms to apply force on the surface of the human body or a certain acupuncture point, and gradually press down with force to achieve the pressing process. It is suitable for flat parts of the human body such as the back, abdomen, and thighs; point: use the protruding part of the flexed knuckles as the point of force to press on a certain treatment point. It has the characteristics of concentrated force and strong stimulation. It is suitable for all parts of the human body; massage: attach the palm or the index, middle, ring, and little finger to the treatment area, and use the wrist joint and forearm to make clockwise or counterclockwise circular movements and friction. It has the effects of accelerating blood circulation and improving tissue metabolism. It is suitable for the human body's waist and back and other parts; push: use the palm of the hand to apply force on a certain part, and perform unidirectional linear push. It is suitable for all parts of the body.

[0028] In the process of demonstrating the above rehabilitation therapy skills, the movement position, posture, speed, force / torque and other information of the rehabilitation robot end are collected. The demonstration trajectory includes position Posture q∈SO(3), speed Force / torque The multi-dimensional motion features collected are aligned in time, space and frequency domains by spline interpolation and other methods, and the aligned motion features are processed by data dimensionality reduction, outlier removal, data smoothing, etc. to form a rehabilitation motion data set, which is expressed as Where s is a high-dimensional input variable, ξ is a high-dimensional output variable, T is the time length after alignment, M is the number of training samples, and the rehabilitation movement database is demonstrated multiple times and continuously improved. The rehabilitation movement database includes movement position-velocity dataset, position-posture dataset, and position-force / torque dataset.

[0029] Specifically, the skill learning steps include:

[0030] Based on the theory of robot skill learning, the Gaussian mixture model-Gaussian mixture regression model (GMM-GMR) is used to construct the rehabilitation robot's motion position-velocity skill model f(v|x), position-posture skill model f(q|x) and position-force / torque skill model f(τ|x) for the motion position-velocity dataset, position-posture dataset and position-force / torque dataset, respectively, to achieve the precise transfer of motion skills from "rehabilitation therapist to rehabilitation robot".

[0031] Assume that the rehabilitation skill position-speed dataset is Gaussian mixture process (GMM) is used to obtain the joint probability distribution P(x,v) of the motion trajectory, and N reference motion trajectories representing the skill input distribution are sampled from GMM. To achieve the empirical modeling of the rehabilitation skill demonstration position of the rehabilitation therapist. Then the Gaussian mixture regression model (GMR) is used to calculate the different reference position trajectories x n The speed corresponding to the input , in order to achieve empirical modeling of the rehabilitation therapist's demonstration speed, i.e.

[0032] Aiming at the demand for posture transformation of rehabilitation robots during rehabilitation therapy, a position-posture skill model f(q|x) is established. The posture information is transformed into Euclidean space through logarithmic mapping λ=log(q) by adopting the space change method. Then, the position-posture skill model is modeled and learned by the joint input of high-dimensional vectors. The position-posture dataset is converted into Convert to Euclidean space In the same way as the position-velocity skill model, the skill model f(λ|x) of position and intermediate variables is constructed by inputting the unknown variable x * Get the mean of the corresponding variable λ, and finally restore the mean of the intermediate variable from the Euclidean space to the desired posture through exponential mapping, f(q|x * )=exp(E(λ(x * ))).

[0033] In view of the demand for force / torque of the rehabilitation robot during rehabilitation therapy, a position-force / torque skill model f(τ|x) is established. It is assumed that the ideal motion force and torque trajectory of the rehabilitation robot terminal is obtained through skill modeling and learning as F d , which can be represented by the F of the virtual mass spring-damper system d Expressed as K P is the unknown stiffness matrix, K D is an unknown damping matrix. Since the expected force of the expected motion trajectory is related to the stiffness matrix, a position-stiffness skill model f(K|x) is established for force control. Perform vectorization and convert it into Euclidean space, and then train the position-stiffness skill modeling and learning through high-dimensional joint input. Similar to the position-velocity skill modeling method, construct the position-stiffness γ skill model f(γ|x) to obtain any unknown position x * As the input, the mean value of the motion trajectory of the corresponding variable γ is restored from the Euclidean space to the stiffness matrix f(K|x * ), and finally realize the conversion of stiffness value to desired force at the end, thereby realizing impedance control.

[0034] Existing rehabilitation therapy robots are restricted by the task space during the learning and reproduction of rehabilitation skills. Skill teaching and skill reproduction are mostly in the same task space. For different patients, rehabilitation skills need to be re-taught according to the differences of the patients themselves, and they lack generalization capabilities. For this reason, in the present invention, after the skill learning step, the skill adjustment step is entered, and the skills of the rehabilitation robot are regulated by combining skill transfer and skill generalization. Specifically, this skill transfer can make the implementation process of rehabilitation skills not restricted by the task space. In the preparation stage of the rehabilitation therapy robot performing the rehabilitation therapy task, the patient's area to be treated is restricted, and the conversion relationship between the restricted area and the robot's base coordinate system is obtained. To prevent the robot from exceeding the treatment range during the treatment task and causing harm to the patient. The three-dimensional offset of the robot end position Δp(t) is obtained through the original teaching trajectory, Δp(t)=p(k+1)-p(k), t represents time, p(k) represents the three-dimensional position information of the kth point in the trajectory, and the end posture based on the robot base coordinates can be expressed as To obtain the transformation relationship between the restricted area and the rehabilitation robot base coordinate system, where R P is the robot posture during the teaching process. The desired position of the rehabilitation therapy robot in the treatment space can be expressed as in R t Robot posture in the human-robot collaboration process.

[0035] Skill generalization is achieved by dynamic motion primitives, which can generalize the rehabilitation skill trajectory to a certain extent according to the individual characteristics of the patients themselves. Dynamic motion primitives (DMPs) can be regarded as a second-order dynamic system with self-stability, that is, Where y is the system state, represents the speed of the system, represents the acceleration of the system, g is the target state, α y , β y is a constant coefficient, σ is a scaling term used to change the convergence speed of the trajectory, and f is a nonlinear function used to change the shape of the trajectory. Ψ i is the Gaussian basis function, ω i is the weight of each basis function, N is the number of basis functions, and the scaling of the original teaching skill trajectory by the nonlinear function f is changed according to factors such as the patient's actual body shape, and the convergence of the skill trajectory from any starting point to different end points is achieved.

[0036] In addition, the existing rehabilitation therapy robot rehabilitation therapy mode generally requires the rehabilitation therapist to set the rehabilitation therapy movement trajectory according to the patient's condition. After completing the rehabilitation therapy process, the rehabilitation therapy movement trajectory is modified according to the patient's treatment condition. It is difficult for the rehabilitation therapist to intervene in the rehabilitation therapy robot in real time based on the patient's real-time feedback during the treatment process. To address the above problems, the rehabilitation robot stability and compliance skill learning and interactive control method provided in this embodiment also includes a human-machine interactive control step, specifically including:

[0037] During rehabilitation therapy, the six-dimensional force sensor on the robot's end reads the interaction force. When external forces interfere with the robot's rehabilitation therapy task and reach a set threshold, the robot enters an interactive control state. In this interactive control state, the rehabilitation therapist can adjust the original rehabilitation skill trajectory based on the patient's treatment feedback to meet the patient's existing treatment needs. In this interactive control state, the rehabilitation therapist drags the robot's end, while compensating for gravity, to modify the original rehabilitation therapy trajectory. During the dragging teaching process, the rehabilitation therapy robot collects information such as the robot's end position, posture, velocity, and force / torque. This information is used to model the rehabilitation skill based on the robot's skill learning method, replacing the corresponding data in the original rehabilitation trajectory and achieving interactive human-machine control. During the interactive control process, the control area is restricted by the rehabilitation therapy area set for skill transfer, ensuring the safety of the control process.

[0038] It can be seen that the stability and compliance skill learning and interactive control method of the rehabilitation robot can respond promptly to the patient's feedback during the rehabilitation therapy process by introducing human-computer interactive control steps. The rehabilitation therapist intervenes in the rehabilitation robot according to the feedback and modifies the rehabilitation therapy motion trajectory.

[0039] Specifically, the variable impedance control step includes:

[0040] During the rehabilitation robot's execution of rehabilitation tasks, the impedance of the rehabilitation robot is adjusted online by setting the task accuracy time-varying impedance robot controller, so as to better adapt to the rehabilitation therapy tasks on the human body surface. The dynamic model of the rehabilitation robot is set as q, are the joint position, velocity and acceleration of the rehabilitation robot, M(q), G(q) is the inertia matrix, Coriolis term and gravity vector respectively. The proportional differential (PD) controller of the rehabilitation robot can be expressed as K J (t),D J (t) are stiffness matrix and damping matrix respectively, τ ext is the torque generated by the external force, τ is the total torque, qd are the desired robot joint positions.

[0041] According to the rehabilitation therapy task trajectory, the task trajectory error value is obtained through Obtain the trajectory error Δx required for calculation ee ,in is the mission trajectory error, represents the initial task error, which is a constant, d(t) actual is the Euclidean distance between the current end position of the rehabilitation robot and the end point of the task trajectory, d(t) initial Represents the Euclidean distance between the initial point and the end point of the trajectory. The six-dimensional force sensor at the end of the rehabilitation robot is used to obtain the interaction force at the end of the robot during the rehabilitation therapy task. Introducing Hooke's law, Among them, F represents the interaction force at the end of the robot, K represents the calculated stiffness value, and the current stiffness value is obtained by the task trajectory error value and the interaction force at the end of the rehabilitation robot, thereby realizing the time-varying impedance control of the rehabilitation robot.

[0042] In addition, in order to ensure the stability of the rehabilitation robot control system, the rehabilitation robot stability compliance skill learning and interactive control method provided in this embodiment also includes:

[0043] The stability of the task accuracy variable stiffness method is guaranteed by introducing the parameterized quadratic Lyapunov function. ,in Represents joint position error, joint velocity error, joint acceleration error, and stiffness matrix K respectively. J (t) and the damping matrix D J (t) is a time-varying matrix. In order to ensure the stability of the system, the derivative of the Lyapunov function must be guaranteed to be semi-negative definite.

[0044] The Lyapunov function is defined as

[0045] T represents the transpose of the vector, and M represents the inertia matrix described above.

[0046] The derivative of the Lyapunov function is Through the rehabilitation robot kinematics and PD controller, the function derivative can be expressed as:

[0047] To meet the above requirements

[0048] To ensure the stability of the system, the stiffness and damping parameters cannot be adjusted arbitrarily. According to the properties of the quadratic form, δ min (A)||x|| 2 ≤x TAx≤δ max (A)||x|| 2 , the relationship between stiffness and damping can be expressed as set up If the damping is less than the stable value, set the damping to I is the identity matrix,

[0049] In this way, according to the above process, the stability of the entire robot control system is achieved

[0050] In summary, this paper constructs a comprehensive rehabilitation skill library for skill demonstration. Regarding skill learning, it constructs the rehabilitation robot's motion position-velocity skill model f(v|x), position-posture skill model f(q|x), and position-force / torque skill model f(τ|x) through skill learning methods. Regarding skill adjustment, it uses skill transfer, dynamic motion primitives (DMPs), and interactive control methods to achieve the generalization of the rehabilitation robot's rehabilitation therapy motion trajectory, as well as the ability for patients, rehabilitation therapists, and rehabilitation robots to interact with each other during the rehabilitation therapy process. Finally, a time-varying impedance controller is designed to ensure the compliant control of the rehabilitation robot, and the stability of the entire control system is ensured through the Lyapunov function.

[0051] The above embodiments are intended only to illustrate the technical concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent changes or modifications made based on the essence of the present invention are intended to be covered by the scope of protection of the present invention.

Claims

1. A method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot, characterized in that: The method comprises: Acquire rehabilitation physiotherapy movement data during the rehabilitation physiotherapy skill demonstration to build a rehabilitation movement database; Constructing a motion position-velocity skill model, a position-posture skill model, and a position-force / torque skill model for the rehabilitation robot based on the rehabilitation motion database to enable the rehabilitation robot to learn rehabilitation therapy skills; Combining skill transfer and skill generalization, the rehabilitation robot is regulated after learning rehabilitation therapy skills to prevent the rehabilitation robot from exceeding the treatment scope during the treatment task and to enable the rehabilitation robot to adapt to different rehabilitation therapy needs; Through the motion trajectory error of the rehabilitation robot and the real-time interactive force feedback of the rehabilitation robot during rehabilitation therapy, the time-varying impedance control of the rehabilitation robot is realized; The combination of skill transfer and skill regulation of the rehabilitation robot after learning rehabilitation therapy skills to prevent the rehabilitation robot from exceeding the scope of treatment during the treatment task process includes: During the preparation stage of the rehabilitation robot's rehabilitation therapy task, the patient's treatment area is restricted and the conversion relationship between the restricted area and the rehabilitation robot's base coordinate system is obtained. The three-dimensional offset of the end position of the rehabilitation robot is obtained through the original teaching trajectory, Δp(t), Δp(t) = p(k+1)-p(k), t represents time, p(k) represents the three-dimensional position information of the kth point in the trajectory, and p(k+1) represents the three-dimensional position information of the k+1th point in the trajectory; the end posture based on the base coordinates of the rehabilitation robot is expressed as Obtain the transformation relationship between the restricted area and the rehabilitation robot base coordinate system, where R P The posture of the rehabilitation robot during the teaching process; The desired position of the rehabilitation robot in the treatment space is expressed as: represents the expected position of the rehabilitation robot end in the robot base coordinate system at time t, represents the expected position of the rehabilitation robot end in the robot base coordinate system at time t-1, where R t Rehabilitation robot posture in human-robot collaboration.

2. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 1, wherein: The method also includes: during the rehabilitation therapy process, when external force interferes with the rehabilitation therapy task of the rehabilitation robot and reaches a set threshold, entering an interactive control state; in the interactive control state, adjusting the motion trajectory of the original rehabilitation skill according to the patient's treatment feedback, and synchronously collecting rehabilitation therapy movement data to update the rehabilitation movement database.

3. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 1, wherein: The method further comprises: introducing a Lyapunov function to perform global stable motion estimation and parameter optimization.

4. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 1, wherein: The rehabilitation therapy movement data during the rehabilitation therapy skill demonstration is obtained through the following methods: A sensor is installed at the end of the rehabilitation robot to measure the interaction force between the rehabilitation robot and the environment in real time; when the rehabilitation robot is gravity compensated, the rehabilitation therapist conducts human-computer interaction with the rehabilitation robot and demonstrates rehabilitation therapy skills to the rehabilitation robot. During the demonstration of rehabilitation therapy skills, motion data of the end of the rehabilitation robot is collected; the collected motion data is processed to form a rehabilitation motion database.

5. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 1 or 4, characterized in that: The rehabilitation exercise database includes a movement position-velocity data set, a position-posture data set, and a position-force / torque data set; The Gaussian mixture model-Gaussian mixture regression model is used to construct the rehabilitation robot's motion position-velocity skill model, position-posture skill model and position-force / torque skill model for the motion position-velocity dataset, position-posture dataset and position-force / torque dataset respectively.

6. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 1, wherein: The skill generalization is achieved by dynamic motion primitives to generalize the rehabilitation skill motion trajectory to a certain extent according to the individual characteristics of the patients themselves.

7. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 1, wherein The dynamic motion primitive is regarded as a second-order dynamic system with self-stability, that is, Where y is the system state, represents the speed of the system, represents the acceleration of the system, g is the target state, α y , β y is a constant coefficient, σ is a scaling term used to change the convergence speed of the trajectory, and f is a nonlinear function used to change the shape of the trajectory; Ψ i is the Gaussian basis function, ω i is the weight of each basis function, and N is the number of basis functions. According to patient factors, the scaling of the original teaching skill trajectory by the nonlinear function f is changed, and the convergence of the skill trajectory from any starting point to different end points is achieved.

8. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 3, wherein: The time-varying impedance control of the rehabilitation robot is achieved by using the motion trajectory error of the rehabilitation robot and the real-time interactive force feedback of the rehabilitation robot during rehabilitation therapy, including: The dynamic model of the rehabilitation robot is set as q, are the joint position, velocity and acceleration of the rehabilitation robot, M(q), G(q) are the inertia matrix, Coriolis term and gravity vector respectively; the proportional differential controller of the rehabilitation robot is expressed as K J (t),D J (t) are stiffness matrix and damping matrix respectively, τ ext is the torque generated by the external force, τ is the total torque, q d is the desired robot joint position; According to the rehabilitation therapy task trajectory, the task trajectory error value is obtained through Obtain the trajectory error Δx required for calculation ee ,in is the mission trajectory error, represents the initial task error, which is a constant, d(t) actual is the Euclidean distance between the current end position of the rehabilitation robot and the end point of the task trajectory, d(t) initial It represents the Euclidean distance between the initial point of the trajectory and the end point of the trajectory. The interaction force of the rehabilitation robot terminal during the rehabilitation therapy task is obtained through the rehabilitation robot terminal sensor. Introducing Hooke's law, Among them, F represents the interaction force at the end of the robot, K represents the calculated stiffness value, and the current stiffness value is obtained by combining the task trajectory error value with the interaction force at the end of the rehabilitation robot to realize the time-varying impedance control of the rehabilitation robot.

9. The method for learning and interactively controlling the stability and compliance skills of a rehabilitation robot according to claim 8, wherein: The introduction of the Lyapunov function for global stable motion estimation and parameter optimization includes: Introducing error variables in Represents joint position error, joint velocity error, joint acceleration error, and stiffness matrix K respectively. J (t) and the damping matrix D J (t) is a time-varying matrix; the derivative of the Lyapunov function is negative semidefinite: The Lyapunov function is defined as: T represents the transpose of the vector, and M represents the inertia matrix; The derivative of the Lyapunov function is: Through the kinematic model of the rehabilitation robot and the proportional differential controller, the Lyapunov function derivative is expressed as: To meet the above requirements According to the properties of quadratic forms, δ min (A)||x|| 2 ≤x T Ax≤δ max (A)||x|| 2 , δ min (),δ max () represent the minimum and maximum values ​​respectively. The relationship between stiffness and damping is expressed as: set up If the damping is less than the stable value, set the damping to I is the identity matrix,

Images