Iterative learning impedance control algorithm for flexible driven exoskeleton

Abstract

The application provides an iterative learning impedance control algorithm for a flexible driving exoskeleton, and the algorithm comprises the following steps: a dynamics model of an upper limb rehabilitation robot is established by measuring parameters such as the length, mass and stiffness of a series elastic actuator of an exoskeleton joint; secondly, the model is processed by using singular perturbation theory to simplify the complexity of the system; on this basis, a human-computer interaction control strategy is designed by combining impedance control and iterative learning algorithm, and collaborative rehabilitation training of the machine active and human active modes is realized; finally, the effectiveness of the algorithm is verified through simulation and experiment platform, and the results show that the method can adapt to the flexible driving characteristics, and improve the tracking accuracy and human-computer interaction safety of the exoskeleton.

Description

Technical Field

[0001] The present invention relates to the field of rehabilitation robots, and in particular to an iterative learning impedance control algorithm of a rehabilitation robot based on a flexible driver. Background Art

[0002] With the advent of an aging population, the number of people with physical disabilities caused by diseases like stroke is rapidly increasing. Traditional rehabilitation models face challenges such as a shortage of rehabilitation trainers, limited training, and long rehabilitation cycles. Rehabilitation robots, as intelligent rehabilitation devices integrating mechanics, electronics, control, and sensing technologies, can provide stroke patients with personalized treatment plans, precise force support, diverse rehabilitation training, and real-time feedback monitoring, significantly reducing the burden on medical staff and improving the efficiency and quality of medical services. However, during rehabilitation training, human-machine interaction is inevitable, which can easily cause secondary injuries to patients and significantly reduce the effectiveness of rehabilitation.

[0003] To address the above problems, the present invention uses a series elastic driver as the driving component of the exoskeleton robot, and combines impedance control with iterative learning algorithm to design a control algorithm. By measuring the interaction force between the robot and the human, the impedance control parameters can be dynamically changed, thereby changing the output torque of the flexible driver, effectively reducing the human-machine confrontation during rehabilitation training. Summary of the Invention

[0004] This paper proposes an iterative learning impedance control algorithm for a flexible drive exoskeleton. By measuring parameters such as the exoskeleton's joint length and mass, as well as the stiffness of the series elastic actuators, a dynamic model of the upper-limb exoskeleton robot is established. Singular perturbation theory is then used to refine the model and simplify the system complexity. Based on this, a human-robot interaction control strategy is designed by combining impedance control with the iterative learning algorithm. Finally, the algorithm's effectiveness is verified through simulation and experimental platforms. Results demonstrate that this method can adapt to the characteristics of flexible drive, improving the exoskeleton's tracking accuracy and the safety of human-robot interaction.

[0005] With reference to the accompanying drawings, the present invention is implemented through the following technical solutions:

[0006] An iterative learning impedance control algorithm for a flexible drive exoskeleton, the control method is as follows:

[0007] Step 1: Measure the joint length, joint mass, and mass and stiffness of the exoskeleton robot's series elastic actuator.

[0008] Step 2: Construct a dynamic model of the upper limb exoskeleton robot based on series elastic actuators.

[0009] Step 2.1: Establish the dynamic model using the Lagrangian method. The specific formula is as follows:

[0010] ,

[0011] in, is the output torque, is the end position of the exoskeleton robot, is the Lagrangian, defined as , is the total kinetic energy of the system, is the total potential energy of the system, and the dynamic model is established as follows:

[0012] ,

[0013] in, is the output torque of the series elastic actuator, is the stiffness matrix of the series elastic actuator, is the motor angular displacement, is the end position of the exoskeleton robot, is the robot terminal speed, is the acceleration of the robot end. and They are inertia matrix, Collio matrix and gravity torque respectively. The specific formulas are:

[0014] ,

[0015] ,

[0016] ,

[0017] in is the moment of inertia of the exoskeleton joint, For the end of the exoskeleton robot, is the length of the upper limb exoskeleton, is the mass of the upper limb exoskeleton joints, is the gravitational acceleration term.

[0018] Step 2.2: Construct the dynamic equation based on the series elastic actuator as:

[0019] ,

[0020] in, is the output torque of the motor, is the inertia matrix of the motor, is the robot end position, is the motor angular displacement, is the second-order derivative of the motor angular displacement.

[0021] Step 3: Based on the singular perturbation theory, the upper limb exoskeleton robot model and the series elastic actuator model are processed and divided into two subsystems, fast and slow.

[0022] Step 3.1: The exoskeleton robot driven by the series elastic actuator exhibits a significant dual time scale characteristic: slow exoskeleton body dynamics and fast actuator elastic dynamics. Based on this, the system can be decoupled by singular perturbation theory, the fast dynamics can be regarded as the perturbation of the slow dynamics, and the control input Designed as two sub-variables: fast and slow and The sum of , is a constant matrix, and the simplified dynamic equation is:

[0023] ,

[0024] In the kinetic equation, is the stiffness of the series elastic actuator, is the inertia matrix of the series elastic actuator, represents the angular displacement of the motor shaft, is the motor angular velocity, is the second-order derivative of the motor angular displacement.

[0025] Step 3.2: Define the joint torque provided by the series elastic actuator , simplify the model to get:

[0026] ,

[0027] Introducing small parameters Establish the singular perturbation equation:

[0028] ,

[0029] in, , , and For both drive stiffness and small parameters Related vectors.

[0030] Step 3.3: When the parameter is small is 0, the driver dynamics equation can be solved :

[0031] ,

[0032] Substitute it into the robot dynamics equation to obtain the overall motion control equation:

[0033] ,

[0034] Finally, the control law can be designed Control the exoskeleton robot.

[0035] Step 4: Combine the impedance control algorithm and the iterative learning algorithm to construct a human-computer interaction control algorithm so that the flexible drive exoskeleton can help subjects complete rehabilitation training.

[0036] Step 4.1: Design the control law It can be expressed by the sum of the following three terms, namely, the stability term , impedance term and base items The stability term is used to ensure system stability, the impedance term can achieve flexible control, and the basis term is the dynamic compensation term in the control law. The control law formula is as follows:

[0037] ,

[0038] in, , , is the error between the expected trajectory and the actual trajectory, is the error between the expected speed and the actual speed, For time, and are the inertia matrix, the Collio matrix and the gravity matrix respectively, is the desired trajectory of the joint, is the desired velocity of the joint, and are the stiffness and damping matrices obtained through iterative learning, is the time-varying iterative learning feedforward term, is a stable term.

[0039] Step 4.2: Design the impedance controller as , the corresponding expression is as follows:

[0040] ,

[0041] in, is the error between the expected trajectory and the actual trajectory, and are the stiffness matrix and damping matrix obtained through iterative learning, For the time-varying iterative learning feedforward term, the three parameter expressions are as follows:

[0042] ,

[0043] ,

[0044] ,

[0045] in, , is a constant, T is the period of each iteration, , and are all symmetric positive definite constant matrices, is the forgetting factor obtained by iterative learning, and the formula is as follows:

[0046] ,

[0047] Among them, the evaluation factor It is used to dynamically adjust the forgetting factor and impedance parameters. The evaluation factor determination formula is as follows:

[0048] ,

[0049] in, and are two constants, To design the weight factor according to the position error, its expression is as follows:

[0050] ,

[0051] in, is a constant between 0 and 1, is the cost function, and its specific expression is as follows:

[0052] ,

[0053] in, is a constant between 0 and 1, For the The expected trajectory of each joint, For the The actual trajectory of each joint.

[0054] Step 4.3: Design the stability part , the corresponding expression is as follows:

[0055] ,

[0056] in, , is the initial robust gain, is the update factor obtained by iterative learning. The specific formula is as follows:

[0057] ,

[0058] Among them, the evaluation factor It is used to dynamically adjust the update factor and thus control the weight of the stability term. The determination formula is as follows:

[0059] ,

[0060] in, is a constant, is the weight factor designed according to the human-computer interaction force, and its expression is as follows:

[0061] ,

[0062] in, is a constant ranging from 0 to 1, where The cost function is expressed as follows:

[0063] ,

[0064] in, is a constant between 0 and 1, For the Human-computer interaction force of each joint.

[0065] Step 5: Conduct simulation verification and platform experiments on the iterative learning impedance control algorithm proposed above for flexible driven exoskeleton to verify the reliability of the algorithm.

[0066] Compared with the prior art, the advantages of the present invention are:

[0067] The present invention constructs an exoskeleton dynamic model of an upper limb rehabilitation robot based on a series elastic driver, and processes the robot model and the driver model based on the singular perturbation theory. Combining impedance control and iterative learning, a human-machine interaction control algorithm is constructed, which has the following characteristics: First, the present invention adopts the singular perturbation theory to process the dynamic model of the flexible driver and the upper limb exoskeleton, which simplifies the robot dynamic equation. Second, the present invention constructs a control strategy based on impedance control and iterative learning by introducing human-machine interaction variables, realizes the change of impedance parameters according to the human-machine interaction force and error, and adjusts the input torque in real time, effectively reducing the secondary injury of the patient and greatly improving the rehabilitation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The overall control block diagram

[0069] Figure 2 Trajectory tracking for simulation trajectory

[0070] Figure 3 Simulated position tracking error

[0071] Figure 4 Simulation update factor

[0072] Figure 5 Simulation forgetting factor

[0073] Figure 6 Platform trajectory tracking

[0074] Figure 7 is the platform position error

[0075] Figure 8 Platform forgetting factor

[0076] Figure 9 Update factors for platform experiments DETAILED DESCRIPTION

[0077] The technical solutions in the embodiments of the present invention are described clearly and completely below in conjunction with the accompanying drawings in the present invention.

[0078] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0079] The present invention is achieved through the following technical solutions:

[0080] Step 1: Measure the joint length and joint mass of the exoskeleton robot and the mass and stiffness of the series elastic actuator.

[0081] The measured length of the boom connecting rod is m, mass is kg, moment of inertia , the inertia of the series elastic actuator is , the spring stiffness is .

[0082] Step 2: Construct the exoskeleton dynamic model of the upper limb rehabilitation robot.

[0083] Step 2.1: Establish the dynamic model by Lagrangian method. The specific formula is as follows:

[0084] ,

[0085] in, is the output torque, For the end of the exoskeleton robot, is the Lagrangian, defined as , is the total kinetic energy of the system, is the total potential energy of the system, and the dynamic model is established as follows:

[0086] ,

[0087] in, is the output torque of the series elastic actuator, is the stiffness matrix of the series elastic actuator, is the motor angular displacement, is the end position of the exoskeleton robot, is the robot terminal speed, is the acceleration of the robot end. and They are inertia matrix, Collio matrix and gravity torque respectively. The specific formulas are:

[0088] ,

[0089] ,

[0090] ,

[0091] in, is the output torque of the motor, is the inertia matrix of the motor, is the robot end position, is the motor angular displacement, is the second-order derivative of the motor angular displacement.

[0092] Step 2.2: Construct the dynamic equation based on the series elastic actuator as:

[0093] ,

[0094] in, is the output torque of the motor, is the inertia matrix of the motor.

[0095] Step 3: Based on the singular perturbation theory, the upper limb exoskeleton robot model and the series elastic actuator model are processed and divided into two subsystems: fast and slow.

[0096] Step 3.1: The exoskeleton robot driven by the series elastic actuator exhibits a significant dual time scale characteristic: slow exoskeleton body dynamics and fast actuator elastic dynamics. Based on this, the system can be decoupled by singular perturbation theory, the fast dynamics can be regarded as the perturbation of the slow dynamics, and the control input Designed as two sub-variables: fast and slow and The sum of , simplifying the dynamic equation to obtain:

[0097] ,

[0098] In the kinetic equation, is the stiffness of the series elastic actuator, is the inertia matrix of the series elastic actuator, represents the joint angle, Indicates the angular displacement of the motor shaft.

[0099] Step 3.2: Define the joint torque provided by the series elastic actuator , simplify the model to get:

[0100] ,

[0101] Introducing small parameters Establish the singular perturbation equation:

[0102] ,

[0103] in , , are two vectors related to the actuator stiffness and small variables.

[0104] Step 3.3: When the parameter is small is 0, the driver dynamics equation can be solved :

[0105] ,

[0106] It can be brought into the robot dynamics equation to obtain the overall motion control equation:

[0107] ,

[0108] in, and They are inertia matrix, Collio matrix and gravity matrix respectively, and finally the control law can be designed Control the robot.

[0109] Step 4: Combining the impedance control algorithm with the iterative learning algorithm, a human-computer interaction control algorithm is constructed to enable the flexible drive exoskeleton to help the subjects complete rehabilitation training;

[0110] Step 4.1: Considering the stability term, the designed control law is:

[0111] ,

[0112] in, , , is the error between the expected trajectory and the actual trajectory, is the error between the expected speed and the actual speed, For time, and are the inertia matrix, the Collio matrix and the gravity matrix respectively, is the desired trajectory of the joint, is the desired velocity of the joint, and are the stiffness and damping matrices obtained through iterative learning, is the time-varying iterative learning feedforward term, is a stable term.

[0113] Step 4.2: Design the impedance controller as , the corresponding expression is as follows:

[0114] ,

[0115] in, is the error between the expected trajectory and the actual trajectory, and are the stiffness and damping matrices obtained through iterative learning, For the time-varying iterative learning feedforward term, the three parameter expressions are as follows:

[0116] ,

[0117] ,

[0118] ,

[0119] in, , is a constant, , and are all symmetric positive definite constant matrices, is the forgetting factor obtained by iterative learning, and the formula is as follows:

[0120] ,

[0121] Among them, the evaluation factor It can be dynamically adjusted according to the real-time changes in human-computer interaction force to achieve adaptive adjustment of impedance control parameters. The determination formula is as follows:

[0122] ,

[0123] in, and are two constants, To design the weight factor according to the position error, its expression is as follows:

[0124] ,

[0125] where, is a constant between 0 and 1, is the cost function, which is expressed as follows:

[0126] ,

[0127] where, is a constant between 0 and 1, is the desired trajectory of the th joint, is the actual trajectory of the th joint.

[0128] Step 4.3: Designing the stabilizing term part , the corresponding expression is as follows:

[0129] ,

[0130] where, , is the initial robust gain, is the update factor obtained by iterative learning, and the specific formula is as follows:

[0131] ,

[0132] where, the evaluation factor is used to dynamically adjust the update factor and thus control the weight of the stabilizing term, and its judgment formula is as follows:

[0133] ,

[0134] where, is a constant, is a weight factor designed according to human-robot interaction force, and its expression is as follows:

[0135] ,

[0136] where, is a constant between 0 and 1, is the cost function, which is expressed as follows:

[0137] ,

[0138] where, is a constant between 0 and 1, is the human-robot interaction force of the th joint.

[0139] Step 5: Simulate and verify the above proposed iterative learning impedance control algorithm for flexible driven exoskeleton, and conduct platform experiments to verify the reliability of the algorithm.

[0140] Figure 2-Figure 5 It is the control algorithm simulation trajectory tracking diagram and the corresponding parameter change diagram. Figure 6-Figure 9 It is the platform experiment trajectory tracking diagram and the corresponding parameter change diagram.

Claims

1. An iterative learning impedance control algorithm for a flexible drive exoskeleton, characterized in that: The following steps are involved: Step 1: Measure the joint length, joint mass, and mass and stiffness of the exoskeleton robot; Step 2: Construct a dynamic model of the upper limb exoskeleton robot based on a series elastic actuator; Step 3: Based on the singular perturbation theory, the upper limb exoskeleton robot model and the series elastic actuator model are processed and divided into two subsystems: fast and slow. Step 4: Combining the impedance control algorithm with the iterative learning algorithm, a human-computer interaction control algorithm is constructed to enable the flexible drive exoskeleton to help the subject complete rehabilitation training. The algorithm of step 4 is specifically implemented as follows: Step 4.1: Design the control law u r It can be expressed by the sum of the following three terms, namely the stability term u stable , impedance term u imp and base term u base The stability term is used to ensure system stability, the impedance term can achieve flexible control, and the basis term is the dynamic compensation term in the control law. The control law formula is as follows: in, e is the error between the expected trajectory and the actual trajectory, is the error between the expected speed and the actual speed, t is the time variable, M(q), and G(q) are the inertia matrix, Collio matrix and gravity matrix respectively, q d is the desired trajectory of the joint, is the desired velocity of the joint, K(t) and D(t) are the stiffness and damping matrices obtained through iterative learning, τ(t) is the time-varying iterative learning feedforward term, and w i ε is the stabilization term; Step 4.2: Design the impedance term u in the control law imp , the corresponding expression is as follows: Among them, the update law of K(t), D(t) and τ(t) through iterative learning can be expressed by the following three expressions: K(t)=K(tT)+N k (εe T -(1+γ i )K(t)), τ(t)=τ(tT)+N τ (e-(1+c) i )τ(t)), Among them, ε is a variable, T is the cycle of iterative learning, defined as λ is a constant, N k ,N D and N τ is a symmetric positive definite constant matrix, γ i is the forgetting factor obtained by iterative learning, and its expression is as follows: c i =(1+ζ)(1+γ) i-1 )-1, Among them, the evaluation factor ζ can be dynamically adjusted according to the real-time changes of the human-computer interaction force to achieve adaptive adjustment of the impedance control parameters; Step 4.3: Design the stabilizing part of the control law u stable , the corresponding expression is as follows: you stable =w i eh, Among them, w i =(1+η)w0, w0 is the initial robust gain, η i is the update factor obtained by iterative learning. The specific formula is as follows: or i =(1+σ)(1+η) i-1 )-1, Among them, the position evaluation factor σ can be dynamically adjusted according to the position error to achieve adaptive adjustment of the forgetting factor; Step 5: Conduct simulation verification and platform experiments on the iterative learning impedance control algorithm proposed above for flexible driven exoskeleton to verify the reliability of the algorithm.

Images