On-demand auxiliary interaction control method for upper limb rehabilitation robot based on potential energy field constraint

Through the control method based on potential energy field constraints, the problems of restricted movement and insufficient participation of upper limb rehabilitation robots are solved, and safe and stable movement and efficient rehabilitation training are achieved in three-dimensional space.

CN119015090AActive Publication Date: 2024-11-26NINGBO INST OF MATERIALS TECH & ENG CHINESE ACAD OF SCI +2

Patent Information

Application Number
CN202410934716.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-26
Estimated Expiration
2044-07-12

AI Technical Summary

Technical Problem

The existing upper limb rehabilitation robots have problems in their designs with limited movement and inability to change rehabilitation tasks according to patient needs, resulting in poor effectiveness in patients' lack of active participation and neuroplasticity during the rehabilitation process.

Method used

Using the on-demand assisted interaction control method of upper limb rehabilitation robot based on potential field constraints, normal and tangential assistance is provided by constructing a three-degree of freedom operation robotic arm and end execution handle, combining positive kinematics model, potential field calculation and RBF network. Strength to achieve safe and stable movement of patients in three-dimensional space.

Benefits of technology

It improves the safety and efficiency of rehabilitation training, enhances the active participation of patients, and ensures the effective completion of rehabilitation tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119015090A_ABST
    Figure CN119015090A_ABST
Patent Text Reader

Abstract

The invention relates to an on-demand auxiliary interaction control method for an upper limb rehabilitation robot based on potential energy field constraint, and the method comprises the steps: building a three-degree-of-freedom operation upper limb rehabilitation robot, presetting a target motion track, building a task space, building a virtual potential energy field based on rehabilitation demands in the task space, and carrying out the compliance control strategy based on the potential energy field. After a patient deviates from a target motion trail in the training process, a potential energy field can provide normal auxiliary force, the effect of restraining unexpected normal vibration on the training trail is achieved in combination with tangential auxiliary force, and the effect of assisting patients with different motion abilities in training according to needs is achieved by adopting an RBF network based on a greedy algorithm. The safe, stable and controllable movement of the upper limb rehabilitation robot in a three-dimensional space is realized; the safety of the patient in the rehabilitation training process is effectively improved, and it is guaranteed that the patient can independently and effectively complete the training task.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of robot control technology, and in particular to an on-demand auxiliary interactive control method for an upper limb rehabilitation robot based on potential energy field constraints. Background Art

[0002] In recent years, with the increasing number of patients with brain injuries and neurodegenerative diseases, especially stroke patients, 80% will experience motor dysfunction, of which 45% will experience loss of upper limb function, including musculoskeletal paralysis. Stroke has become the third highest cause of death in China. In the field of rehabilitation, therapists will repeatedly bend the affected limbs of patients to reduce muscle spasms and stimulate neuroplasticity, so that the brain can form new neural connections. This traditional method is time-consuming, laborious and costly. In addition, there is a serious shortage of rehabilitation therapists in my country at this stage, and the rehabilitation efficiency is low, which cannot meet the increasing rehabilitation needs. Therefore, rehabilitation-assisted robots have emerged as a new type of rehabilitation-assisted medical equipment. However, for physical therapy, whether assisted by therapists or robots, there is a limitation, because the patient's movement guidance is completely provided from the outside, so the patient has almost no participation in the entire treatment process. This involuntary rehabilitation task has been pointed out to be worthless. In the case of acute stroke, the first 3 to 6 months of rehabilitation almost determines the patient's final level of paralysis, and psychological problems are a very important factor during rehabilitation. Therefore, the patient's active participation in training is necessary for the effectiveness of rehabilitation. In recent years, mirror therapy and imagery training have been shown to stimulate brain and neural plasticity, stimulate different areas of the patient's brain to produce movement, sensation, and pain awareness, and induce the patient's active participation. These therapies have been applied and practiced around the world.

[0003] The upper limb rehabilitation robot in the prior art simulates the rehabilitation techniques of the rehabilitation therapist and repeatedly bends the arms and limbs. However, this mirror tensioner can only control the input force of the upper limb on one side through the mirror image input force of the other side. It only uses the replication of the mirror image force as its training goal, and does not control the input force. When the input force on one side is too large, the tensioner on the other side will output a larger acceleration or move to a farther distance, which is easy to strain the user's upper limb. The above reasons are mainly the following two points:

[0004] (1) Most rehabilitation robots are designed with a two-degree-of-freedom (DOF) master / slave configuration to produce elbow flexion / extension and forearm extension / flexion. However, the end of this type of robot can only move in a two-dimensional plane and cannot provide three-dimensional spatial movement for the patient's upper limbs. The activity space is limited, and the rehabilitation task cannot be changed according to the patient's needs.

[0005] (2) Passive training in upper limb rehabilitation can stimulate the patient's motor plasticity, thereby improving the patient's motor ability. However, this passive training strictly follows the target trajectory set by the therapist, completely disobeys the patient's motor intention, and even moves against the patient's intention. Obviously, the patient will gradually lose active participation in passive rehabilitation training. Passive training is very ineffective in stimulating the patient's neural plasticity. Summary of the invention

[0006] The technical problem to be solved by the present invention is how to achieve safe, stable and controllable movement of an upper limb rehabilitation robot in three-dimensional space. In order to overcome the defects of the above-mentioned prior art, the present invention provides an interactive control method of an upper limb rehabilitation robot based on potential energy field constraints.

[0007] The present invention provides an on-demand assisted interactive control method for an upper limb rehabilitation robot based on potential energy field constraints, which is applied to an upper limb rehabilitation robot operating with three degrees of freedom. The upper limb rehabilitation robot comprises a mechanical arm with three rotational joints and an end-actuator handle. The three rotational joints of the mechanical arm are driven to rotate by three drive motors. The three rotational joints are provided with angle encoders for collecting the rotation angles of the rotational joints. A force sensor for collecting force data applied by a patient to the end-actuator handle during human-machine interaction is provided at the end-actuator handle. The on-demand assisted interactive control method for an upper limb rehabilitation robot comprises:

[0008] Step 1: Preset the target motion trajectory of the upper limb rehabilitation robot and evenly sample N sampling points from the target motion trajectory: In the formula, It is represented as the three-dimensional column vector of the end effector handle of the upper limb rehabilitation robot at the i-th point in the Cartesian space; D p A discrete data set representing the target motion trajectory of the end effector handle;

[0009] Step 2: Establish a forward kinematics model of the upper limb rehabilitation robot through the rotation angles of the three rotation joints of the robotic arm, and calculate the actual position of the end effector handle in the reference coordinate system {0} based on the forward kinematics model;

[0010] Step 3: Set the task space based on the target motion trajectory and calculate the sampling points in the task space The total potential energy of other points p in the task space; and design the negative gradient of the potential energy - Preset the mapping factor δ and combine the mapping factor δ with the negative gradient - Calculate the normal auxiliary force F in the direction of motion of the end-effector handle p ;

[0011] Step 4: Calculate gravity compensation and friction compensation based on the position of the end effector handle on the reference coordinate system {0};

[0012] Step 5: Calculate the tangential auxiliary force F in the running direction of the end effector handle based on the position of the end effector handle in the reference coordinate system {0} and the target motion trajectory. i ;

[0013] Step 6: Based on the actual position of the end effector handle calculated in step 2, a projection module is established to facilitate the operation of the RBF network in three-dimensional space. The projection ε of the end effector handle in the constraint direction is calculated by the projection module. The projection ε and the force data collected during the human-computer interaction process are input into the Gaussian RBF network to iteratively calculate the maximum force F that the patient can apply to the end effector handle. m ; Preset the auxiliary coefficient α of the patient's force, and according to the maximum force F m , tangential auxiliary force F i and α are used to calculate the weighted tangential auxiliary force F r ;

[0014] Step 7: Based on the normal auxiliary force F p and weighted tangential auxiliary force F r Calculate the compensation force F of the end effector handle;

[0015] Step 8: Based on the gravity compensation and friction compensation in step 4 and the compensation force F and Jacobian matrix J calculated in step 7 T Calculate the torque mapped to each rotational joint of the upper limb rehabilitation robot.

[0016] Compared with the prior art, the present application has the following advantages: based on constructing an upper limb rehabilitation robot operating in three degrees of freedom, presetting a target motion trajectory and building a task space, constructing a virtual potential field based on rehabilitation needs in the task space, and based on a compliant control strategy of the potential field, when the patient deviates from the target motion trajectory during training, the potential field will provide a normal auxiliary force, which is combined with the tangential auxiliary force to achieve the effect of constraining the undesired normal vibration on the training trajectory, and adopts an RBF network based on a greedy algorithm to achieve the effect of assisting the training of patients with different motor abilities on demand, thereby realizing safe, stable and controllable movement of the upper limb rehabilitation robot in three-dimensional space; effectively improving the safety of patients during rehabilitation training, and ensuring that patients can independently and effectively complete training tasks.

[0017] In a possible implementation, the step 2 of calculating the actual position of the end effector handle on the reference coordinate system {0} based on the forward kinematics model includes:

[0018] p x =a2cθ1cθ2+(-d2-d4)sθ1+a3cθ1cθ2cθ3-a3cθ1sθ2sθ3

[0019] p y=(d2+d4)cθ1+a2sθ1cθ2+a3sθ1cθ2cθ3-a3sθ1sθ2sθ3

[0020] p z =-a3sθ 23 -a2sθ2

[0021] Where θ1, θ2, and θ3 are the angle values ​​of the three rotating joints of the robot arm measured by the sensor, and a i represents the length of the common normal between the rotation axis of the i-th revolute joint and the rotation axis of the adjacent i+1-th revolute joint; d i represents the common normal of the ith revolute joint and the adjacent ith+1th revolute joint and d i represents the distance between the common normal of the ith revolute joint and the i-1th revolute joint along the rotation axis of the ith revolute joint, cθ i is cos(θ i ), sθ i is sin(θ i ), sθ ij is sin(θ i +θ j ), cθ ij is cos(θ i +θ j ) is a simplified form of .

[0022] In a possible implementation, in step 3, the task space is set according to the target motion trajectory, and the sampling points in the task space are calculated. The total potential energy of other points p in the task space includes:

[0023] Step 301A: According to Hooke's law, calculate the sampling points in the task space The potential energy at each point p is:

[0024]

[0025] In the formula, Represents each sampling point The initial potential energy, and for each sampling point The initial potential energy The optimization is performed by solving the convex optimization problem, which can be expressed as and γ i The constraints are Point p and sampling point There is a virtual spring between the sampling points The attraction to point p is K irepresents the stiffness. After integration, the elastic potential energy of point p is

[0026] Step 302A: Calculate the potential energy weight w from point p to N sampling points using the Gaussian kernel function i (p):

[0027]

[0028] In the formula, σ i is the weight of the influence range of each sampling point;

[0029] Step 303A: i (p) Normalized to get

[0030]

[0031] Step 304A: i (p) Multiply and sum to get the total potential energy of each point p in the task space

[0032]

[0033] The negative gradient of the design potential energy in step 3 is - It is expressed as:

[0034]

[0035] In step 3, the mapping factor δ is preset, and the mapping factor δ is combined with the negative gradient - Calculate the normal auxiliary force F in the direction of motion of the end-effector handle p , expressed as:

[0036] In a possible implementation manner, the calculation of gravity compensation and friction compensation based on the position of the end effector handle on the reference coordinate system {0} in step 4 specifically includes:

[0037] The calculation of gravity compensation specifically includes:

[0038] Calculate the torque of a revolute joint:

[0039]

[0040] Substituting the robot arm parameters into the torque calculation formula of the rotary joint, we get:

[0041]

[0042]

[0043] The gravity compensations of the three revolute joints are -τ1, -τ2, and -τ3 respectively;

[0044] The calculation method of friction compensation is as follows: collect the current of the three rotating joints of the upper limb rehabilitation robot running at a constant speed and the data relationship between the speed and friction obtained by changing the running speed; introduce the speed and friction data into the Stribeck model to obtain a fitting model with speed and friction as independent variables, namely friction compensation, where the Stribeck model is: Among them, a, b, c, d, n are the parameters to be fitted, v and f represent the speed and friction in the fitted data respectively, and e represents the natural logarithm.

[0045] In a possible implementation manner, the tangential auxiliary force F in the running direction of the end effector handle is calculated in step 5. i Specifically include:

[0046] Calculate the position difference Δp=x between the actual position x and the target position of the upper limb rehabilitation robot d -x;

[0047] The position difference Δp is input into the impedance controller to calculate the tangential auxiliary force F in the running direction of the end-actuator handle. i :

[0048]

[0049] In the formula, Represents the differential of potential difference; K s represents the stiffness of the impedance controller, B d represents the damping of the impedance controller; where, In the formula, d represents the distance of the target motion trajectory; one round trip along the target motion trajectory is recorded as a cycle, and T is the one-way cycle; x d It is converted into a displacement relative to the starting point in a predefined training direction to generate the target position in three-dimensional space.

[0050] In a possible implementation, the projection of the end effector handle in the constraint direction is calculated by a projection module in step 6, specifically including: the projection module obtains the starting position and the ending position of the end effector handle based on the starting angle and the ending angle of the end effector handle, obtains a vector V1 based on the starting position and the ending position, obtains a vector V2 based on the difference between the current actual position of the end effector handle and the starting position, and obtains the projection ε of the end effector handle in the constraint direction by performing the vector inner product of V1 and V2;

[0051] The projection ε and the force data collected during the human-computer interaction are input into the Gaussian RBF network to iteratively calculate the maximum force F that the patient can apply to the end effector handle.m , including:

[0052] F m =Φ T ω

[0053] Φ=[g1g2g3……g n ] T

[0054] Where Φ is an n×1 vector containing radial basis functions, ω is a weight vector containing n parameters to be estimated; n is the number of RBFs distributed on the straight line in the task space; g i represents the i-th Gaussian radial basis function, g i The expression is: In the formula, μ i Uniformly distributed in the task space, represents the center of the i-th Gaussian RBF; x represents the current actual position; σ represents the scalar constant of the width of the Gaussian radial basis function;

[0055] The upper limb end robot is trained without potential field constraints to collect kinematic data and force sensor measurement data, and the initial weight vector of the RBF network based on the least squares method is determined by the kinematic data and force sensor measurement data:

[0056]

[0057] F=[F s1 F s2 F s3 ...F sm ] T

[0058] ω0=(H T H) -1 H T F

[0059] Where H is the n×m matrix containing the time series of the Gaussian RBF network, m represents the mth moment, and n represents the number of nodes in the Gaussian RBF network; F sm represents the measurement data of the force sensor of the end effector handle at the mth moment, (T T h) -1 H T is the pseudo-inverse of the matrix H; ω0 is the initial weight vector of the Gaussian RBF network;

[0060] In the iterative training of the Gaussian RBF network, the weight vector ω of the Gaussian RBF network is updated based on the initial weight vector ω0 using the gradient descent algorithm, specifically including:

[0061]

[0062] ωk+1 =ω k +Δω

[0063] In the formula, λ represents the learning rate, t represents the sampling point in the task cycle, k represents the task amount, Δω is the average gradient over a task cycle, and ω k and ω k+1 are the weight vectors of the current task cycle and the next task cycle respectively;

[0064] In step 6, according to the prediction force F m , tangential auxiliary force F i and the auxiliary coefficient α are used to calculate the weighted tangential auxiliary force F r The calculation formula is:

[0065] F r =F i -αF m .

[0066] In a possible implementation manner, the compensation force F of the end effector handle in step 7 is F p +F r .

[0067] In a possible implementation, the calculation formula of the Jacobian matrix J in step 8 is:

[0068]

[0069] The calculation formula for calculating the torque ψ mapped to each rotation joint of the upper limb rehabilitation robot in step 8 is:

[0070] ψ=J(q) T *F

[0071] Where J(q) T is a 3×3 Jacobian matrix, F is a 3×1 matrix, and ψ is a 3×1 matrix. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a structural diagram and coordinate system distribution diagram of the three-degree-of-freedom rehabilitation robot of the present invention;

[0073] Figure 2 It is a system block diagram of the on-demand assisted interactive control method of the upper limb rehabilitation robot based on potential energy field constraint of the present invention;

[0074] Figure 3 To determine the trend of the subjects’ average force with respect to the number of tasks, the average force of all subjects under different tasks was calculated, and a first-order linear function was used to fit the relationship between the average force and the number of tasks. DETAILED DESCRIPTION

[0075] First, those skilled in the art should understand that these implementations are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0076] In the description of the embodiments of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the embodiments of the present application can be understood according to specific circumstances.

[0077] In the embodiments of the present application, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0078] The present application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0079] See also Figure 1 to Figure 3 As shown, the embodiment of the present application discloses an on-demand assisted interactive control method for an upper limb rehabilitation robot based on potential energy field constraints, which is applied to an upper limb rehabilitation robot operating with three degrees of freedom. The upper limb rehabilitation robot has three rotary joints, three drive motors for assisting the rotation of the rotary joints on demand, and an end-actuator handle. The three rotary joints are provided with angle encoders for collecting the rotation angles of the rotary joints. The end-actuator handle is provided with a force sensor for collecting force data applied by the patient to the end-actuator handle during human-machine interaction. The on-demand assisted interactive control method for the upper limb rehabilitation robot includes:

[0080] Step 1: For the patient's rehabilitation training, preset the target motion trajectory of the upper limb rehabilitation robot, and evenly sample N sampling points from the target motion trajectory: In the formula, It is represented as the three-dimensional column vector of the end effector handle of the upper limb rehabilitation robot at the i-th point in the Cartesian space; D p A discrete data set representing the target motion trajectory of the end effector handle;

[0081] Step 2: Establish the forward kinematics model of the upper limb rehabilitation robot through the rotation angles of the three rotation joints of the robotic arm, and calculate the actual position of the end effector handle in the reference coordinate system {0} based on the forward kinematics model, as shown in Figure 1 As shown, in this specific embodiment, the reference coordinate system {0} coincides with the coordinate system {1} at the first rotation joint of the upper limb rehabilitation robot, the second rotation joint is set with a coordinate system {2}, and the third rotation joint is set with a coordinate system {3}; specifically including:

[0082] p x =a2cθ1cθ2+(-d2-d4)sθ1+a3cθ1cθ2cθ3-a3cθ1sθ2sθ3

[0083] p y =(d2+d4)cθ1+a2sθ1cθ2+a3sθ1cθ2cθ3-a3sθ1sθ2sθ3

[0084] p z =-a3sθ 23 -a2sθ2

[0085] Where θ1, θ2, and θ3 are the angle values ​​of the three rotating joints of the robot arm measured by the sensor, and a i It represents the length of the common normal between the rotation axis of the i-th revolute joint and the rotation axis of the adjacent i+1-th revolute joint, that is, the length of the common normal along the x in the coordinate system. i Direction, from z i-1 to z i Translation distance; d i represents the common normal of the ith revolute joint and the adjacent ith+1th revolute joint and d i It represents the distance between the common normal of the i-th revolute joint and the i-1-th revolute joint along the rotation axis of the i-th revolute joint, that is, along the z axis in the coordinate system. i Direction, from x i-1 to x i The distance of translation; cθ i is cos(θ i ), sθ i is sin(θ i ), sθ ij is sin(θ i +θ j ), cθ ij is cos(θ i +θ j )

[0086] Step 3: Set the task space based on the target motion trajectory and calculate the sampling points in the task space The total potential energy of other points p in the task space; and design the negative gradient of the potential energy - Preset the mapping factor δ and combine the mapping factor δ with the negative gradient - Calculate the normal auxiliary force F in the direction of motion of the end-effector handle p ; Among them, about the sampling points in the calculation task space The total potential energy for other points p in the task space specifically includes:

[0087] Step 301A: According to Hooke's law, calculate the sampling points in the task space The potential energy at each point p is:

[0088]

[0089] In the formula, Represents each sampling point The initial potential energy, and for each sampling point The initial potential energy The optimization is performed by solving the convex optimization problem, which can be expressed as and γ i The constraints are Point p and sampling point There is a virtual spring between them. Represented as sampling point Attraction to point p, K i Represents stiffness. The higher the stiffness, the stronger the constraint effect of the potential energy field. After integration, the elastic potential energy of point p is

[0090] Step 302A: Calculate the potential energy weight w from point p to N sampling points using the Gaussian kernel function i (p):

[0091]

[0092] In the formula, σ i is the weight of the influence range of each sampling point;

[0093] Step 303A: i (p) Normalized to get

[0094]

[0095] Step 304A: i (p) Multiply and sum to get the total potential energy of each point p in the task space

[0096]

[0097] About the negative gradient of the design potential - It is expressed as:

[0098]

[0099] About calculating the normal auxiliary force F in the motion direction of the end effector handle p , specifically including: presetting the mapping factor δ, and combining the mapping factor δ with the negative gradient - Calculate the normal auxiliary force F in the direction of motion of the end-effector handle p , the calculation formula is expressed as:

[0100] Step 4: Calculate gravity compensation and friction compensation based on the position of the end effector handle on the reference coordinate system {0}; specifically, the calculation of gravity compensation specifically includes:

[0101] Calculate the torque of a revolute joint:

[0102]

[0103] In the formula, is the inertia, is the Coriolis force and centripetal force, G i , i is the gravity term;

[0104] Substituting the robot arm parameters into the torque calculation formula of the rotary joint, we get:

[0105]

[0106]

[0107] The gravity compensations of the three revolute joints are -τ1, -τ2, and -τ3 respectively;

[0108] The calculation method of friction compensation is as follows: collect the current of the driving motors of the three rotating joints of the upper limb rehabilitation robot running at a constant speed and the data relationship between the speed and friction obtained by changing the running speed; introduce the speed and friction data into the Stribeck model to obtain a fitting model with speed and friction as independent variables, namely friction compensation, where the Stribeck model is: Among them, a, b, c, d, n are the parameters to be fitted, v, f represent the speed and friction in the fitted data respectively, and e represents the natural logarithm;

[0109] Step 5: Calculate the tangential auxiliary force F in the running direction of the end effector handle based on the position of the end effector handle in the reference coordinate system {0} and the target motion trajectory. i ; Specifically include:

[0110] Calculate the position difference Δp=x between the actual position x and the target position of the upper limb rehabilitation robot d -x;

[0111] The position difference Δp is input into the impedance controller to calculate the tangential auxiliary force F in the running direction of the end-actuator handle. i :

[0112]

[0113] In the formula, Represents the differential of potential difference; K s represents the stiffness of the impedance controller, B d represents the damping of the impedance controller; where, In the formula, d represents the distance of the target motion trajectory; one round trip along the target motion trajectory is recorded as a cycle, and T is the one-way cycle; x d Convert it into displacement relative to the starting point in the predefined training direction to generate the target position in three-dimensional space;

[0114] Step 6: Based on the actual position of the end effector handle calculated in step 2, a projection module is established to facilitate the operation of the RBF network in three-dimensional space. The projection ε of the end effector handle in the constraint direction is calculated by the projection module. The projection ε and the force data collected during the human-computer interaction process are input into the Gaussian RBF network to iteratively calculate the maximum force F that the patient can apply to the end effector handle. m ; Preset the auxiliary coefficient α of the patient's force, and according to the maximum force F m , tangential auxiliary force F i and α are used to calculate the weighted tangential auxiliary force F r ;

[0115] The method of calculating the projection ε of the end-effector handle in the constraint direction by the projection module specifically includes: the projection module obtains the starting position and the ending position of the end-effector handle based on the starting angle and the ending angle of the end-effector handle, obtains a vector V1 based on the starting position and the ending position, obtains a vector V2 based on the difference between the current actual position of the end-effector handle and the starting position, and obtains the projection ε of the end-effector handle in the constraint direction by performing the vector inner product of V1 and V2;

[0116] The projection ε and the force data collected during the human-computer interaction are input into the Gaussian RBF network to iteratively calculate the maximum force F that the patient can apply to the end effector handle. m , including:

[0117] F m =Φ T ω

[0118] Φ=[g1g2g3......g n ] T

[0119] Wherein, Φ is an n×1 vector containing radial basis functions, ω is a weight vector containing n parameters to be estimated; n is the number of RBFs distributed on the straight line in the task space. In this specific embodiment, in order to balance the accuracy of the approximation and the computational complexity of the upper limb rehabilitation robot, the number of RBFs is set to 20; g i represents the i-th Gaussian radial basis function, g i The expression is: In the formula, μ i Evenly distributed in the task space, represents the center of the i-th Gaussian RBF; x represents the actual current position of the patient; σ represents the scalar constant of the width of the Gaussian radial basis function. In this specific embodiment, d represents the distance between the centers of the first and last Gaussian radial odd functions;

[0120] In this specific embodiment, the Gaussian RBF network fits the maximum force applied by the patient that varies with time and is used to model the patient's ability; the calculation of the weight vector parameters of the Gaussian RBF network includes an initialization part and an iteration part; since the update of the Gaussian RBF network is based on the weight vector of the RBF network of the previous task, it is necessary to obtain the initial weight vector; specifically, the method includes: training the upper limb end robot to collect kinematic data and measurement data of the force sensor without potential field constraints, and determining the initial weight vector of the RBF network based on the least squares method through the kinematic data and measurement data of the force sensor:

[0121]

[0122] F=[F s1 F s2 F s2 ...F sm ] T

[0123] ω0=(H T H) -1 H T F

[0124] Where H is the n×m matrix containing the time series of the Gaussian RBF network, m represents the mth moment, and n represents the number of nodes in the Gaussian RBF network; F sm represents the measurement data of the force sensor of the end effector handle at the mth moment, (H T h) -1 H T is the pseudo-inverse of the matrix H; ω0 is the initial weight vector of the Gaussian RBF network;

[0125] After obtaining the initial weight vector ω0, in the iterative training of the Gaussian RBF network, the weight vector ω of the Gaussian RBF network is updated based on the initial weight vector ω0 using a gradient descent algorithm, specifically including:

[0126]

[0127] ω k+1 =ω k +Δω

[0128] In the formula, λ represents the learning rate, t represents the sampling point in the task cycle, k represents the task amount, Δω is the average gradient over a task cycle, and ω k and ω k+1 are the weight vectors of the current task cycle and the next task cycle respectively; where L t is a binary variable whose value is obtained by comparing the measured value of the force sensor with the estimated value of the Gaussian RBF network. st Greater than F mt , it means that the output of the Gaussian RBF network is less than the maximum force that the patient can exert. Therefore, the weight vector needs to be updated to increase the output of the Gaussian RBF network. On the contrary, when F st Less than F mt , this means that the patient should have been able to exert more force; therefore, the control will maintain the weight vector to challenge the patient and stimulate them to exert more active force; since the output of the Gaussian RBF network is only updated towards an increasing trend, the proposed robot assistance strategy is called a greedy on-demand assistance (GAAN) controller. In addition, if the patient occasionally reduces the force they exert due to other subjective reasons such as slackness, the controller will not misjudge their functional ability to decline. Therefore, the greedy on-demand assistance control strategy in this specific embodiment establishes a patient's motor ability model based on the maximum force exerted by the patient during rehabilitation training rather than the instantaneous force;

[0129] According to the prediction power F m , tangential auxiliary force F i and the auxiliary coefficient α are used to calculate the weighted tangential auxiliary force F r The calculation formula is:

[0130] F r =F i -αF m ;

[0131] Step 7: Based on the normal auxiliary force F p and weighted tangential auxiliary force F r Calculate the compensation force F of the end effector handle; the specific calculation formula is: F = F p +F r ;

[0132] Step 8, based on the gravity compensation and friction compensation in step 4 and the compensation force F and Jacobian matrix J calculated in step 7, calculate the torque mapped to each rotation joint of the upper limb rehabilitation robot;

[0133] Among them, the calculation formula of the Jacobian matrix J is:

[0134]

[0135] The calculation formula for calculating the torque ψ mapped to each rotation joint of the upper limb rehabilitation robot in step 8 is:

[0136] ψ=J(q) T *F

[0137] Where J(q) T is a 3×3 Jacobian matrix, F is a 3×1 matrix, and ψ is a 3×1 matrix.

[0138] Experimental methods:

[0139] The experiment was completed by 6 healthy volunteers (4 males and 2 females), aged 24.5±0.6 years and 173.0±7.2cm in height. All experiments were approved by the Ethics Committee of Ningbo Institute of Materials Technology and Engineering, Chinese Academy of Sciences. All subjects provided informed consent. During the experiment, the volunteers lay on the bed-ridden stroke patient synchronous training rehabilitation robot.

[0140] Each subject was asked to perform three sets of rehabilitation training exercises. Before the training began, we showed the subjects the virtual environment and explained the operations, and gave them a few minutes to familiarize themselves with the system. During the experiment, the subjects completed the shooting task under the guidance of visual feedback, and the amount of force applied by the subjects was determined by the subjects' own will. When the subject reached the current target, the virtual man fired a bullet to break the target ball, and the subject began to move towards the starting point. The target ball was regenerated 2 seconds after reaching the starting point. The cycle of each shooting task was set to 10s, and the kinematic and kinetic data were recorded at a frequency of 1000Hz during the experiment.

[0141] The experiment aims to verify that the developed on-demand assistance controller based on potential field constraints can promote the active participation of subjects, in which each subject performs three sets of training exercises with different assistance levels (α=0, 0.5, 1) under the same impedance controller. The initial weight vectors of the RBF network are set to the values ​​obtained by each subject in the previous experiment. In the training exercise with α=0, the weight vector of the RBF network is updated normally based on the method of 4.2.2, but the output of the network does not affect the robot controller. Therefore, the control paradigm with α=0 cannot provide challenges for the subjects, while the controllers of the other groups (α=0.5, 1) can use the RBF network to provide challenges for the subjects. Each set of training includes 10 back and forth movements. Before the next new exercise starts, the weight vector of the RBF network is updated according to the acquired data. The iteration step λ is set to 0.2.

[0142] In order to further analyze the relationship between the subjects' participation and the task cycle at the same challenge level, the number of tasks was used as the independent variable, and the data obtained are shown in the figure:

[0143] Figure 3 The average force change curve of the subject as the number of tasks is performed and the fitting curve based on the first-order linear function is depicted. The slope of the fitting curve represents the changing trend of the subject's participation during the task; Figure 3 (a), (b), and (c) represent the motion results of endpoint AB when α = 0, α = 0.5, and α = 1 respectively; Figure 3 (d) (e) (f) respectively represent the motion results of endpoint BA when α = 0, α = 0.5, and α = 1

[0144] .Depend on Figure 3 (a)(b)(c) It can be seen that when α=0, the force of the subject in each task cycle is almost unchanged (slope=-0.005). When α=0.5 or α=1, as the number of tasks increases, the average force of the subject also increases (slope=0.323, slope=0.484).

[0145] Figure 3 Similar conclusions can be drawn for the reverse motion recorded in (d), (e), and (f), which indicates that the task challenge provided by the proposed controller forced the subjects to provide greater active force to complete the training task, effectively avoiding slackness.

[0146] In the description of the embodiments of the present application, it should be noted that in the description of the present application, terms such as "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is only for the convenience of description, and does not indicate or imply that the device or component must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present application.

[0147] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" etc. means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0148] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for on-demand assisted interaction control of an upper limb rehabilitation robot based on potential field constraints, applied to an upper limb rehabilitation robot operating in three degrees of freedom, wherein the upper limb rehabilitation robot has three rotary joints, a drive motor for assisting the rotation of the rotary joints when necessary, and an end-actuator handle, wherein the three rotary joints are provided with angle encoders for collecting the rotation angles of the rotary joints, and the end-actuator handle is provided with a force sensor for collecting the force data applied by the patient to the end-actuator handle during human-machine interaction, characterized in that: The on-demand auxiliary interactive control method of the upper limb rehabilitation robot comprises: Step 1: Preset the target motion trajectory of the upper limb rehabilitation robot and evenly sample N sampling points from the target motion trajectory: In the formula, It is represented as the three-dimensional column vector of the end effector handle of the upper limb rehabilitation robot at the i-th point in the Cartesian space; D p a discrete data set representing the target motion trajectory of the end effector; Step 2: Establish a forward kinematics model of the upper limb rehabilitation robot through the rotation angles of the three rotation joints of the robotic arm, and calculate the actual position of the end effector handle in the reference coordinate system {0} based on the forward kinematics model; Step 3: Set the task space based on the target motion trajectory and calculate the sampling points in the task space The total potential energy of other points p in the task space; and design the negative gradient of the potential energy Preset the mapping factor δ and combine the mapping factor δ with the negative gradient Calculate the normal auxiliary force F in the direction of motion of the end-effector handle p ; Step 4: Calculate gravity compensation and friction compensation based on the position of the end effector handle on the reference coordinate system {0}; Step 5: Calculate the tangential auxiliary force F in the running direction of the end effector handle based on the position of the end effector handle in the reference coordinate system {0} and the target motion trajectory. i ; Step 6: Based on the actual position of the end effector handle calculated in step 2, a projection module is established to facilitate the operation of the RBF network in three-dimensional space. The projection ε of the end effector handle in the constraint direction is calculated by the projection module. The projection ε and the force data collected during the human-computer interaction process are input into the Gaussian RBF network to iteratively calculate the maximum force F that the patient can apply to the end effector handle. m ; Preset the auxiliary coefficient α of the patient's force, and according to the maximum force F m , tangential auxiliary force F i and α are used to calculate the weighted tangential auxiliary force F r ; Step 7: Based on the normal auxiliary force F p and weighted tangential auxiliary force F r Calculate the compensation force F of the end effector handle; Step 8: Based on the gravity compensation and friction compensation in step 4 and the compensation force F and Jacobian matrix J calculated in step 7 T Calculate the torque mapped to each rotational joint of the upper limb rehabilitation robot.

2. The upper limb rehabilitation robot interactive control method based on potential energy field constraint according to claim 1 is characterized in that: Calculating the actual position of the end effector handle on the reference coordinate system {0} based on the forward kinematics model in step 2 includes: p x =a2cθ1cθ2+(-d2-d4)sθ1+a3cθ1cθ2cθ3-a3cθ1sθ2sθ3 p v =(d2+d4)cθ1+a2sθ1cθ2+a3sθ1cθ2cθ3-a3sθ1sθ2sθ3 p z =-a3sθ 23 -a2sθ2 Where θ1, θ2, and θ3 are the angle values ​​of the three rotating joints of the robot arm measured by the sensor, and a i represents the length of the common normal between the rotation axis of the i-th revolute joint and the rotation axis of the adjacent i+1-th revolute joint; d i represents the common normal of the ith revolute joint and the adjacent ith+1th revolute joint and d i represents the distance between the common normal of the ith revolute joint and the i-1th revolute joint along the rotation axis of the ith revolute joint, cθ i is cos(θ i ), sθ i is sin(θ i ), sθ ij is sin(θ i +θ j ), cθ ij is cos(θ i +θ j ) is a simplified form of .

3. The upper limb rehabilitation robot interactive control method based on potential energy field constraint according to claim 2 is characterized in that: In step 3, the task space is set according to the target motion trajectory, and the sampling points in the task space are calculated. The total potential energy of other points p in the task space includes: Step 301A: According to Hooke's law, calculate the sampling points in the task space The potential energy at each point p is: In the formula, Represents each sampling point The initial potential energy, and for each sampling point The initial potential energy The optimization is performed by solving the convex optimization problem, which can be expressed as and γ i The constraints are Point p and sampling point There is a virtual spring between the sampling points The attraction to point p is K i represents the stiffness. After integration, the elastic potential energy of point p is Step 302A: Calculate the potential energy weight w from point p to N sampling points using the Gaussian kernel function i (p): In the formula, σ i is the weight of the influence range of each sampling point; Step 303A: i (p) Normalized to get Step 304A: i (p) and Multiply and sum to get the total potential energy of each point p in the task space The negative gradient of the design potential energy in step 3 It is expressed as: In step 3, the mapping factor δ is preset, and the mapping factor δ is combined with the negative gradient Calculate the normal auxiliary force F in the direction of motion of the end-effector handle p , expressed as:

4. The upper limb rehabilitation robot interactive control method based on potential energy field constraint according to claim 3 is characterized in that: The calculation of gravity compensation and friction compensation based on the position of the end effector handle on the reference coordinate system {0} in step 4 specifically includes: The calculation of gravity compensation specifically includes: Calculate the torque of a revolute joint: Substituting the robot arm parameters into the torque calculation formula of the rotary joint, we get: The gravity compensations of the three revolute joints are -τ1, -τ2, and -τ3 respectively; The calculation method of friction compensation is as follows: collect the current of the three rotating joints of the upper limb rehabilitation robot running at a constant speed and the data relationship between the speed and friction obtained by changing the running speed; introduce the speed and friction data into the Stribeck model to obtain a fitting model with speed and friction as independent variables, namely friction compensation, where the Stribeck model is: Among them, a, b, c, d, n are the parameters to be fitted, v and f represent the speed and friction in the fitted data respectively, and e represents the natural logarithm.

5. The upper limb rehabilitation robot interactive control method based on potential energy field constraint according to claim 4 is characterized in that: The tangential auxiliary force F in the running direction of the end effector handle is calculated in step 5. i Specifically include: Calculate the position difference Δp=x between the actual position x and the target position of the upper limb rehabilitation robot d -x; The position difference Δp is input into the impedance controller to calculate the tangential auxiliary force F in the running direction of the end actuator handle. i : In the formula, Represents the differential of potential difference; K s represents the stiffness of the impedance controller, B d represents the damping of the impedance controller; where, In the formula, d represents the distance of the target motion trajectory; one round trip along the target motion trajectory is recorded as a cycle, and T is the one-way cycle; x d It is converted into a displacement relative to the starting point in a predefined training direction to generate the target position in three-dimensional space.

6. The upper limb rehabilitation robot interactive control method based on potential energy field constraint according to claim 5 is characterized in that: The step 6 of calculating the projection ε of the end-effector handle in the constraint direction by the projection module specifically includes: the projection module obtains the starting position and the ending position of the end-effector handle based on the starting angle and the ending angle of the end-effector handle, obtains a vector V1 based on the starting position and the ending position, obtains a vector V2 based on the difference between the current actual position of the end-effector handle and the starting position, and obtains the projection ε of the end-effector handle in the constraint direction by performing the vector inner product of V1 and V2; The projection ε and the force data collected during the human-computer interaction are input into the Gaussian RBF network to iteratively calculate the maximum force F that the patient can apply to the end effector handle. m , specifically including: F m =Φ T oh Φ=[g1 g2 g3......g n ] T Where Φ is an n×1 vector containing radial basis functions, ω is a weight vector containing n parameters to be estimated; n is the number of RBFs distributed on the straight line in the task space; g i represents the i-th Gaussian radial basis function, g i The expression is: In the formula, μ i Uniformly distributed in the task space, represents the center of the i-th Gaussian RBF; x represents the current actual position; σ represents the scalar constant of the width of the Gaussian radial basis function; The upper limb end robot is trained without potential field constraints to collect kinematic data and force sensor measurement data, and the initial weight vector of the RBF network based on the least squares method is determined by the kinematic data and force sensor measurement data: F=[F s1 F s2 F s3 ...F sm ] T ω0=(H T H) -1 H T F Where H is the n×m matrix containing the time series of the Gaussian RBF network, m represents the mth moment, and n represents the number of nodes in the Gaussian RBF network; F sm represents the measurement data of the force sensor of the end effector handle at the mth moment, (H T H) 1 H T is the pseudo-inverse of the matrix H; ω0 is the initial weight vector of the Gaussian RBF network; In the iterative training of the Gaussian RBF network, the weight vector ω of the Gaussian RBF network is updated based on the initial weight vector ω0 using the gradient descent algorithm, specifically including: oh k+1 =ω k +See In the formula, λ represents the learning rate, t represents the sampling point in the task cycle, k represents the task amount, Δω is the average gradient over a task cycle, and ω k and ω k+1 are the weight vectors of the current task cycle and the next task cycle respectively; In step 6, according to the prediction force F m , tangential auxiliary force F i and the auxiliary coefficient α are used to calculate the weighted tangential auxiliary force F r The calculation formula is: F r =F i -αF m 。 7. The upper limb rehabilitation robot interactive control method based on potential energy field constraint according to claim 6 is characterized in that: The compensation force F of the end effector handle in step 7 is F p +F r .

8. The upper limb rehabilitation robot interactive control method based on potential energy field constraint according to claim 7 is characterized in that: The calculation formula of the Jacobian matrix J in step 8 is: The calculation formula for calculating the torque ψ mapped to each rotation joint of the upper limb rehabilitation robot in step 8 is: ψ=J(q) T *F Where J(q) T is a 3×3 Jacobian matrix, F is a 3×1 matrix, and ψ is a 3×1 matrix.

Citation Information

Patent Citations

  • Robot exploration path generation method and device, computer equipment and storage medium

    CN110531759A

  • Control method for passive rehabilitation training of flexible upper limb rehabilitation robot

    CN112237524A

  • Mirror image mechanical arm control method based on potential energy field and mirror image mechanical arm equipment

    CN114434452A

  • Lower limb rehabilitation robot compliance control method

    CN114848391A

  • On-demand auxiliary control method based on lower limb exoskeleton robot

    CN115252369A

Cited By

  • Torque compensation method and system for upper limb rehabilitation robot, storage medium and computer

    CN119626453A

  • Upper limb rehabilitation robot torque compensation method, system, storage medium and computer

    CN119626453B

  • Rehabilitation training anti-loosening method and device based on AAN challenge degree adjustment

    CN120340759A