Trajectory tracking control method for exoskeleton robot, and exoskeleton robot system

Abstract

Embodiments of the present application set forth a trajectory tracking control method for an exoskeleton robot, comprising: in a first stage, when an exoskeleton robot is not applied to a subject, determining system interference of a power system of the exoskeleton robot, and compensating an output parameter of a power model of the exoskeleton robot; in a second stage, when the exoskeleton robot is applied to a subject, isolating the system interference from a human-machine interaction torque of the second stage, and determining the human-machine interaction torque; inputting the human-machine interaction torque into a controller, and using the controller to determine an assistance of the power system of the exoskeleton robot.

Description

Trajectory tracking control method of exoskeleton robot and exoskeleton robot system Technical Field

[0001] The present application relates to a trajectory tracking control method for an exoskeleton robot and an exoskeleton robot system. Background Art

[0002] A typical application of exoskeletons is rehabilitation, where they can assist therapists with labor-intensive training, thereby alleviating the shortage of human resources. The use of exoskeletons for upper limb rehabilitation, including shoulder abduction and extension, shoulder flexion and extension, upper arm internal and external rotation, elbow flexion and extension, and forearm internal and external rotation, is becoming increasingly important.

[0003] To reduce joint inertia and enable more user-friendly training, cable-driven mechanisms have been introduced into upper-limb exoskeletons. This allows the motor to be positioned away from the human limb, minimizing the impact of the machine on the inertia of human motion. In some studies, cable-driven mechanisms use ball screws to reduce system nonlinearity caused by friction. Prior art proposes flexible transmission structures based on Bowden cables, in which the cables are used to transmit force to drive the robot joints. Other studies have designed a rope-driven differential structure that uses two series elastic actuators and ropes to control the rotation and swing of the forearm. Furthermore, some studies have proposed portable and compact cable-driven exoskeleton robots with three degrees of freedom: shoulder abduction and extension, shoulder flexion and extension, and elbow flexion and extension. Considering the geometric calibration of cable-driven upper-limb exoskeletons, researchers have proposed adaptive calibration methods and optimized models of the human-exoskeleton motion system.

[0004] In addition, soft actuators have been used to drive robots, absorb shocks to handle external impact forces, and ensure safe interactions.

[0005] While cable-driven mechanisms and flexible actuators can improve the performance and safety of exoskeleton robots, they make the dynamic model a complex, high-order system that is difficult to calculate. Furthermore, the system's disturbances and the human-robot interaction torque are coupled on the same side of the dynamic model equation.

[0006] For cable-driven exoskeleton robots with flexible actuators, the robot controller should exploit human interaction torque to improve training; at the same time, it should resist unmodeled disturbances caused by the cable-driven mechanism while achieving trajectory tracking accuracy and real-time performance to ensure the effectiveness of training.

[0007] In summary, the advantages of using cable-driven exoskeleton robots with series elastic actuators can be summarized in two aspects: 1) the inertia of the robot joints is relatively low, which is more suitable for human-robot interaction; 2) the elastic elements are tolerant to impact, thus providing structural safety.

[0008] However, existing systems of this type are affected by unmodeled disturbances due to the cable-driven mechanism and external torques due to human-machine interaction, which affect the control accuracy.

[0009] Summary of the Invention

[0010] This application proposes a trajectory tracking control method for an exoskeleton robot and an exoskeleton robot system. The trajectory tracking control method for the exoskeleton robot is divided into two stages: the first stage determines and compensates for the unmodeled disturbance of the cable drive mechanism; the second stage uses the results of the first stage to drive the robot joints to track the desired trajectory.

[0011] The scheme of this application is as follows:

[0012] A trajectory tracking control method for an exoskeleton robot, comprising:

[0013] In a first stage in which the exoskeleton robot is not used by a subject, determining a system disturbance of a power system of the exoskeleton robot and compensating output parameters of a power model of the exoskeleton robot;

[0014] In a second phase of applying the exoskeleton robot to a subject, isolating the system disturbance from the human-machine interaction torque of the second phase and determining the human-machine interaction torque;

[0015] The human-machine interaction torque is input into a controller, and the controller is used to determine the power assistance of the power system of the exoskeleton robot.

[0016] In some optional embodiments, in a first stage where the exoskeleton robot is not used by a subject, the steps of determining the system disturbance of the power system of the exoskeleton robot and compensating the output parameters of the power model of the exoskeleton robot include:

[0017] In the first stage when the exoskeleton robot is not applied to a subject, the system disturbance of the power system of the exoskeleton robot is estimated through iterative learning, and the output parameters of the power model of the exoskeleton robot are compensated.

[0018] In some optional embodiments, the step of estimating the system disturbance of the power system of the exoskeleton robot through iterative learning and compensating the output parameters of the power model of the exoskeleton robot is to estimate the system disturbance of the power system of the exoskeleton robot using an iterative learning impedance controller following the backstepping method.

[0019] In some optional embodiments, multiple joints of the exoskeleton robot are driven by series elastic actuators, and the iterative learning impedance controller is expressed as:

[0020] Formula (1) represents the dynamic equation of the rigid joint side, and Formula (2) represents the dynamic equation of the series elastic actuator side, where q∈R n Represents the joint position vector, θ∈R n Represents the motor rotor shaft position vector of the joint of the exoskeleton robot, M(q)∈R n×n , and g(q)∈R n denote the inertia matrix, centrifugal and Coriolis torques, and gravitational torque of the exoskeleton robot, respectively, K∈R n×n is the stiffness matrix, B∈R n×n is the inertia matrix of the actuator, τ e ∈R n is the physical interaction torque vector, τ f ∈R n represents the disturbance u∈R caused by the cable transmission and joint system friction of the exoskeleton robot n represents the control torque applied to the series elastic actuator of the exoskeleton robot.

[0021] In some optional embodiments, in the step of inputting the human-machine interaction torque into a controller and using the controller to determine the assist force of the power system of the exoskeleton robot, the controller is a suboptimal model predictive controller.

[0022] In some optional embodiments, the suboptimal model predictive controller is a suboptimal model predictive controller following a singular perturbation method.

[0023] In some optional embodiments, the suboptimal model predictive controller includes a relatively slow joint subsystem controller and a relatively fast actuator subsystem controller, and the input of the suboptimal model predictive controller is expressed as: Ⅱ =u f +u s

[0024] Among them, u f and u s represent the control terms of fast and slow time scales, respectively, used to stabilize the joint subsystem and the actuator subsystem.

[0025] The present application also provides an exoskeleton robot system, comprising an exoskeleton robot and a series elastic actuator for driving multiple active joints of the exoskeleton robot; the exoskeleton robot further comprises a processor and a memory, wherein the memory stores computer executable code. When the computer executable code is executed, the processor performs the following operations:

[0026] In a first stage in which the exoskeleton robot is not used by a subject, determining a system disturbance of a power system of the exoskeleton robot and compensating output parameters of a power model of the exoskeleton robot;

[0027] In a second phase of applying the exoskeleton robot to a subject, isolating the system disturbance from the human-machine interaction torque of the second phase and determining the human-machine interaction torque;

[0028] The human-machine interaction torque is input into a controller, and the controller is used to determine the power assistance of the power system of the exoskeleton robot.

[0029] In some optional embodiments, the series elastic execution also includes: a cable, a motor, a first fixed block, a second fixed block, a sleeve, and an elastic element; wherein the tight end of the cable is connected to the sliding part of the motor, the first fixed block can move on the sleeve, the sleeve is fixed on the second fixed block, the loose end of the cable is connected to the joint, when the tight end of the cable moves in the first direction, the first fixed block moves therewith and compresses the elastic element; when the elastic element is compressed, the second fixed block generates a force toward the first direction; the fixed block moves in the first direction along with the loose end of the cable.

[0030] In some optional embodiments, each of the series elastic actuators further includes two potentiometers to measure the compression of the elastic element.

[0031] Compared with other control methods, this control scheme has two advantages:

[0032] First, the present application can separate the coupled disturbance and interaction torque without a dedicated force sensor, thus compensating them in the first stage and utilizing them in the second stage;

[0033] Second, although the overall dynamic model is a high-order system, the order of calculation can be reduced in a reasonable way, so the dynamic model can achieve high accuracy and low complexity at the same time.

[0034] The new trajectory tracking control method for an exoskeleton robot and the exoskeleton robot system proposed in the embodiments of the present application include at least the following effects:

[0035] First, compared with the prior art solutions, the method of the present application does not require a force sensor, which reduces costs. At the same time, it can avoid errors such as noise that are common in force sensors, thereby improving the accuracy of detection and compensation.

[0036] Second, the method of the present application reduces the order of calculation and the amount of calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] FIG1 is a flowchart showing the steps of a trajectory tracking control method for a rope-driven flexible upper limb exoskeleton robot according to an embodiment of the present application.

[0038] FIG2 is a schematic diagram of a rope-driven flexible upper limb exoskeleton robot according to an embodiment of the present application.

[0039] FIG3 is a block diagram of a rope-driven flexible upper limb exoskeleton robot according to an embodiment of the present application.

[0040] FIG4 is a schematic diagram showing a joint of a rope-driven flexible upper limb exoskeleton robot driven by a series elastic actuator (SEA) according to an embodiment of the present application.

[0041] FIG5 shows the control scheme of the second stage of the trajectory tracking control method for the rope-driven flexible upper limb exoskeleton robot.

[0042] Figures 6 and 7 are schematic diagrams of continuous disturbances and time-varying disturbances.

[0043] Figure 8 shows a comparison of the use and non-use of impedance controllers.

[0044] Figures 9 and 10 are snapshots of the first phase at different times and snapshots of the second phase at different times.

[0045] Figure 11 shows the trajectories of joints 1, 2, and 5 of the rope-driven flexible upper limb exoskeleton robot.

[0046] Figure 12 shows the trajectories of joints 1, 2, and 5 based on the proportional-integral-derivative control algorithm.

[0047] Implementation Method

[0048] The embodiments of the present application propose a two-stage trajectory tracking control method for an exoskeleton robot and an exoskeleton robot system, which separates coupled disturbances and interaction torques without dedicated force sensors, thereby compensating for them in the first stage and utilizing them in the second stage.

[0049] FIG1 is a flowchart of the steps of the exoskeleton robot according to an embodiment of the present application. It can be seen that the solution includes the following steps:

[0050] S101, in a first stage where the exoskeleton robot is not used by a subject, determining a system disturbance of a power system of the exoskeleton robot and a corresponding system disturbance compensation value, and compensating output parameters of a power model of the exoskeleton robot using the system disturbance compensation value;

[0051] S102, in the second stage of applying the exoskeleton robot to the subject, separating the system interference from the human-machine interaction torque to determine the human-machine interaction torque;

[0052] S103: Input the human-machine interaction torque into a controller, and use the controller to determine the power assistance of the power system of the exoskeleton robot.

[0053] Among them, the step S103, i.e. calculating the power assist of the exoskeleton robot based on the human-machine interaction torque, specifically includes: inputting the human-machine interaction torque into the second-stage SMPC controller, the controller calculating the appropriate power assist, and adjusting the output power of the power system of the exoskeleton robot.

[0054] FIG2 is a schematic diagram of a flexible upper limb exoskeleton robot for rope traction according to an embodiment of the present application. As shown in FIG2 , the robot is composed of five active joints (J1-J5) and one passive joint (not shown). The six joints correspond to the following movements:

[0055] Joint 1 J1: forearm internal and external rotation; Joint 2 J2: elbow flexion and extension; Joint 3 J3: upper arm internal and external rotation; Joint 4 J4: shoulder flexion and extension; Joint 5 J5: shoulder abduction / adduction; Joint 6 is passive and is used to coordinate eccentric movement of the shoulder joint to break through the limitation of upper limb range of motion and increase the user's exercise training space.

[0056] Figure 3 shows a block diagram of a rope-driven flexible upper limb exoskeleton robot according to one embodiment of the present invention. Each joint is equipped with an encoder (e.g., QY2204-SSI encoders for joints J1 and J3, and 3590S-2-104L encoders for joints J2, J4, and J5) and a harmonic reduction servo motor (AK80-64 servo motors for joints J1-J3, and LSG-32-142-80 servo motors for joints J4 and J5). Furthermore, the embedded mainboard communicates with the slave board via a controller area network (CAN) to drive the five active joints, while the computer communicates with the device via a universal asynchronous receiver / transmitter (UART).

[0057] Figure 4 shows a schematic diagram of the joints of a rope-driven flexible upper-limb exoskeleton robot driven by a series elastic actuator (SEA). As shown in Figure 4, all active joints are driven by a series elastic actuator (SEA). The tight end E1 of the wire rope is connected to a sliding element (e.g., a pulley) at the motor end. A fixed block B1 can move on a sleeve, which is fixed to a fixed block B2. The loose end E2 of the wire rope is connected to the moving joint component. When the tight end E1 of the wire rope moves rightward, fixed block B1 moves rightward with the rope end, compressing elastic element S1 (e.g., a spring). The compression of spring S1 generates a rightward force on fixed block B2. Fixed block B2 moves rightward with the loose end of the wire rope, achieving overall rightward motion. Each series elastic actuator is equipped with two potentiometers to measure spring compression. The difference between the potentiometer values ​​is divided by the pulley radius to obtain the motor shaft angle.

[0058] The following details the first phase in which the exoskeleton robot was not used on the subjects.

[0059] The dynamic model of this robot driven by a series elastic actuator can be expressed as:

[0060] The above formula (1) describes the dynamic equation of the rigid joint side, and the formula (2) describes the dynamic equation of the series elastic actuator side. The above two formulas are coupled by the elastic element elastic force term K(θ-q). Where, q∈R n Represents the joint position vector, θ∈R n Represents the motor rotor shaft position vector, M(q)∈R n×n , and g(q)∈R n They represent the robot’s inertia matrix, the torques generated by centrifugal and Coriolis forces, and the gravitational torque, respectively, K∈R n×n is the stiffness matrix, B∈R n×n is the inertia matrix of the actuator, τ e ∈R n is the physical interaction torque vector, τ f ∈R n Denotes the disturbance u∈R caused by cable transmission and joint system friction n Indicates the control torque applied to the actuator.

[0061] Please note that the overall dynamic model described by the above formulas (1) and (2) is a high-order system, including the rigid joint part corresponding to formula (1) and the flexible actuator part corresponding to formula (2). It is not easy to stabilize and control such a system. In addition, the parameter τ in the above formula is e and τ fare coupled at the rigid joints and cannot be isolated using a single force / torque sensor; however, distinguishing them is crucial to achieving the control goal of isolating the unmodeled disturbance τ f , and using the interaction torque τ e .

[0062] The exoskeleton robot needs to guide the patient through repetitive movements through close interaction to help stroke patients recover their motor function. To achieve this goal, the exoskeleton robot is controlled to track a time-varying trajectory under the desired impedance model:

[0063] where q d ∈R n is the vector of the reference trajectory, M d , C d , K d ∈R n×n Denote the desired inertia matrix, damping matrix, and stiffness matrix, respectively, which are all diagonal and positive definite. Tracking the reference trajectory under the impedance model allows the patient to deviate from the trajectory, thereby providing a certain degree of flexibility to improve safety.

[0064] The above two stages are compared and summarized in Table 1 below.

[0065] Table 1 Comparison of the two control stages

[0066] Specifically, the first stage is an unmodeled perturbation τ to the dynamical system f The first stage is to approximate and compensate; the second stage is to track the desired trajectory according to the desired impedance model (3) to perform rehabilitation training. In this application, the controller used in the two stages does not require a dedicated force sensor and is computationally efficient for high-order systems.

[0067] The following describes the experiments conducted on the two-stage trajectory tracking control method for a rope-driven flexible upper limb exoskeleton robot disclosed in this application:

[0068] First, no human subjects are involved in the first stage, so the above formula (1) can be changed to:

[0069] Next, in the embodiment of the present application, an iterative learning impedance controller is designed for the above formulas (2) and (4). The above method for designing an iterative learning impedance controller can be referenced in the following paper: X. Li, Y. Liu, and H. Yu, “Iterative learning impedance control for rehabilitation robots driven by series elastic actuators,” Automatica, vol. 90, pp. 1–7, 2018.

[0070] The development of this iterative learning impedance controller follows a backstepping approach. The backstepping approach can be found in the following papers: A. Saberi, P. Kokotovic, and H. Sussmann, “Global stabilization of partially linear composite systems,” SIAM Journal on Control and Optimization, vol. 28, no. 6, pp. 1491–1503, 1990; and Y. Pan, H. Wang, X. Li, and H. Yu, “Adaptive command-filtered backstepping control of robot arms with compliant actuators,” IEEE Transactions on Control Systems Technology, vol. 26, no. 3, pp. 1149–1156, 2018.

[0071] By the above method, the above formula (4) can be rewritten as:

[0072] where Δθ = θ - θ d , introduce a desired input vector θ d ∈R n , defined as:

[0073] where K s ∈R n×n is a positive definite diagonal matrix, s∈R n is a sliding vector defined as:

[0074] where α is a positive constant, is defined as the reference vector, m∈R nrepresents a feedforward control vector, which is updated according to the following law: k+1 =m k -βs k (8)

[0075] Where β is a positive constant and the subscript k represents the kth iteration.

[0076] Next, to ensure that the actual state of the actuator converges to the desired input (i.e., θ→θ d ), the actual control input applied to the actuator side can be set as:

[0077] where K θ ∈R n×n is a diagonally positive definite matrix, yes The time derivative of is another reference vector defined as Where γ is a positive constant.

[0078] The controller described by equations (6) and (9) can rewrite the closed-loop system equation as:

[0079] Next, we can prove by deduction that: in steady state, τ f +m→0. In other words, the unmodeled disturbance is well compensated. The derivation process can be referred to the following paper: X.Li, Y.Liu, and H.Yu, “Iterative learning impedance control for rehabilitation robots driven by series elastic actuators,” Automatica, vol. 90, pp. 1–7, 2018.

[0080] In step S101, using the proposed iterative learning controller for disturbance compensation has the following advantages:

[0081] First, it isolates and approximates the disturbance from other dynamic parameters without using any force sensors;

[0082] Second, it exploits the repetitive nature of rehabilitation training to achieve efficient computation; the compensation is feed-forward and updated only within periodic intervals.

[0083] It should be noted that in the above formula (7), the expected time variation trajectory q is further set d The same trajectory was used in the formal rehabilitation training (i.e., the second phase), so that in both phases τf f does not change significantly.

[0084] The following is a detailed description of the second phase of applying the exoskeleton robot to the subjects.

[0085] In the second stage, after the unmodeled disturbances are well compensated, another control scheme is introduced to enable the above-mentioned exoskeleton robot to interact with the subject and perform the trained task, that is, to track the desired trajectory under the desired impedance model.

[0086] The step S102, i.e., the step of isolating the estimated system disturbance and human interaction torque and compensating the output power of the power system in the second stage, is to use a suboptimal model predictive controller to drive the robot to track the desired trajectory.

[0087] The development of the second-stage controller follows the singular perturbation method, which can be understood from the following papers: P.V. Kokotovic, H.K. Khalil, and J. O'Reilly, “Singular perturbation methods in control: analysis and design,” 1986; and X. Li, Y. Pan, G. Chen, and H. Yu, “Multi-modal control scheme for rehabilitation robotic exoskeletons,” The International Journal of Robotics Research, vol. 36, pp. 759–777, 2017.

[0088] In the second stage of the exoskeleton robot control system, the entire system can be decomposed into a slower joint subsystem (the subsystem represented by the above formula 1) and a faster actuator subsystem (the subsystem represented by the above formula 2). All dynamic parameters including disturbances have been obtained in the first stage. Therefore, the overall control input in the second stage can be expressed as u Ⅱ =u f +u s (12)

[0089] Among them, u f and u s They represent the control terms of fast and slow time scales, respectively, and are used to stabilize the two subsystems of the above formula (1) and (the above formula 2). First, u f Designed to:

[0090] where Kv ∈R n×n is a diagonally positive definite matrix.

[0091] Next, we design a model predictive control (MPC) framework to achieve the impedance control task. Specifically, based on formula (1), we define an extended state vector as where z∈R n is the impedance vector, expressed as:

[0092] where τ l Represents τ e The low-pass filtered signal, λ is the M in formula (3) d 、C d , K d The positive constant of the desired impedance parameter setting; it has been proven in the industry that the convergence of z→0 means that the desired impedance mode can be achieved in the low frequency range. In addition, and Respectively represent τ e and τ f Here is the estimated value of is obtained using a nonlinear disturbance observer, and Already in the first phase (i.e. ). Relevant knowledge about nonlinear disturbance observers can be obtained from the following paper: A. Mohammadi, H. J. Marquez, and M. Tavakoli, “Nonlinear disturbance observers: Design and applications to euler?lagrange systems,” IEEE Control Systems, vol. 37, pp. 50–72, 2017.

[0093] Now, the dynamic model of the slower joint subsystem can be described as:

[0094] The control input u s Subject to the following constraints:

[0095] Where U is a set, k is the current moment, and N p is the prediction horizon. The control objective is to ensure that the state vector within the prediction time domain is as close as possible to the reference value. In other words, the cumulative error between the predicted state vector and the reference value should be minimized. In addition, to ensure human safety, the control action should not be too large, and the size of the control vector needs to be constrained.

[0096] Therefore, the cost function of model predictive control (MPC) is defined as:

[0097] Among them, u s (ζ)∈U,ζ∈[t,t+N p ] (18)

[0098] Where Q∈R n×n , R∈R n×n is a symmetric positive definite weighting matrix.

[0099] The control inputs are generated by solving a quadratic optimization problem:

[0100] Specifically, extract U k The first element of the vector is used as the control quantity for this control cycle. In order to ensure the real-time performance of the controller, a maximum number of iterations can be set. When the controller reaches the maximum number of iterations, the solution process stops and a suboptimal solution is generated. In the second stage, the structure of the proposed control scheme (12) is shown in Figure 5. According to Figure 5, the development of the second stage controller follows the singular perturbation method, where u f and u s Represents the control terms for the fast and slow time scales. The second stage controller has the following advantages:

[0101] First, although the overall dynamic model is a high-order system, the control input does not require high-order time derivatives, so there is not much computational complexity; second, although the trajectory control task is achieved under the desired impedance model, the development of the controller does not require force sensors.

[0102] In fact, subsequent experiments demonstrate that the proposed controller can run smoothly even on a microprocessor with very limited resources.

[0103] The following describes the simulation of the two-stage trajectory tracking control method for the exoskeleton robot disclosed in this application.

[0104] A simulation study was conducted to verify the performance of the proposed two-stage control scheme. Specifically, a simulation model of a single-degree-of-freedom cable-driven robot was established by setting the dynamic parameters and the desired impedance parameters in Table 2. In addition, the desired time-varying trajectory was set to q. qx = 0.1sin(0.2π·t-π / 2) (20)

[0105] Table 2 Parameters of the simulation model

[0106] First, the proposed ILC controller (9) is implemented in the first stage according to Algorithm 1, where the disturbance is predefined and set to the ground truth value. Specifically, a continuous disturbance (i.e., τf = 0.02 Nm) and a periodic disturbance (i.e., τf = 0.02 sin(0.2π·t-π / 2) Nm) are sequentially applied to the robot, representing the disturbance from the cable drive mechanism.

[0107] The iterative learning control formula (9) proposed in the first stage is implemented according to the above algorithm flow, in which the interference caused by the rope traction mechanism and friction is predefined and set to true value. Specifically, a constant disturbance (i.e., τ f =0.02Nm) and time-varying disturbances (ie τ f =0.02sin(0.2π·t-π / 2)Nm), which is used to simulate the unmodeled disturbance of the system.

[0108] In each trajectory cycle, a total of N = 100 sampling points are considered, and the value of the slip vector s at each sampling point is recorded; at the end of each cycle, the feedforward term m for the entire cycle is updated according to formula (8), where the update gain is set to 0.3 and 1.0 for continuous disturbances and time-varying disturbances, respectively; in the next cycle, the input θd at each sampling point is added accordingly based on the updated m.

[0109] The results are shown in Figure 6 (for continuous perturbations) and Figure 7 (for time-varying perturbations). Figure 6 (a) shows the results after 3 iterations for continuous perturbations; (b) shows the results after 10 iterations for continuous perturbations; and (c) shows the results after 30 iterations for continuous perturbations. Figure 7 (a) shows the results after 3 iterations for time-varying perturbations; (b) shows the results after 10 iterations for time-varying perturbations; and (c) shows the results after 30 iterations for time-varying perturbations. It can be seen that after 30 iterations, m+τ f →0 converges to 0 (error is less than 0.001Nm).

[0110] Next, the proposed model predictive control (MPC) controller for the second stage is implemented according to Algorithm 2 described below. A comparison is also performed to demonstrate the effectiveness of the first-stage disturbance compensation. As shown in Figure 8, with this compensation, the impedance error (i.e., vector z in (14)) is smaller in the steady state, indicating that the desired impedance model (3) is better implemented.

[0111] The following describes the experiments conducted on the two-stage trajectory tracking control method for the rope-traction flexible-driven upper limb exoskeleton robot disclosed in this application.

[0112] Experiments were conducted to verify the performance of the proposed control scheme by implementing it on joints 1, 2, and 5 of the robot (see Figure 1). The upper limb rehabilitation device consists of an upper limb assembly, an aluminum alloy bracket, and a lifting stool. The aluminum alloy bracket's primary function is to bear the mass of all components; the lifting stool adjusts the subject's sitting height to accommodate the exoskeleton structure. The motor's output torque is transmitted to the joints through the extension and retraction of cables to control their movement.

[0113] In the experiment, the desired trajectory is set to q d :

[0114] First, the first stage is used to estimate and compensate for unmodeled disturbances; thus, only the exoskeleton robot is controlled to follow a predefined trajectory without involving the human subject (i.e., τ e =0), snapshots at different moments can be seen in Figure 9.

[0115] Figure 9 shows snapshots of the first stage at different times: (a) t = 0, the robot’s initial position; (b) t = 3.0 s, joint 2 begins to swing downward, and joint 5 begins to swing outward; (c) t = 5.0 s, joint 2 swings downward to its lowest point, and joint 5 swings outward to its highest point; (d) t = 10.0 s, the robot returns to its initial position.

[0116] Figure 10 shows snapshots of the second stage at different times: (a) t = 0, the robot’s initial position; (b) t = 3.0 s, joint 2 begins to swing downward, and joint 5 begins to swing outward; (c) t = 5.0 s, joint 2 swings downward to its lowest point, and joint 5 swings outward to its highest point; (d) t = 10.0 s, the robot returns to its initial position.

[0117] Then, in the second stage, the disturbance is compensated and the nonlinear disturbance observer (NDOB) is applied to design the impedance control input to control the robot and guide the human body to follow the desired trajectory. f To stabilize the actuator, a slow-time-scale control input is proposed based on the suboptimal model predictive control (SMPC) algorithm to minimize the impedance error. Figure 11 shows the trajectories of joints 1, 2, and 5. (a) Figure 11 shows joint J1, (b) joint J2, and (c) joint J5.

[0118] For comparison, the present application also implements a proportional integral differential (PID) control algorithm to track the same trajectory. Figure 12 shows the trajectory of the proportional integral differential control algorithm. In Figure 12, (a) is joint J1, (b) is joint J2, and (c) is joint J5. Compared with the PID control algorithm, the method proposed in this application ensures a smaller error in the joint (average error: 3.42° vs 1.73°), proving that the proposed control scheme has better performance. The PID algorithm and the two-stage trajectory tracking control scheme proposed in this application are both implemented on a 168MHz single-chip microcomputer (STM32F4). The execution time of each control loop is set to 10 milliseconds. Compared with other control algorithms with similar tracking accuracy, the strategy proposed in this application requires lower hardware cost and sampling frequency. For example, in the paper Y.yao Wang, S. Li, D. Wang, F. Ju, B. Chen, and H. Wu, “Adaptive time-delay control for cable-driven manipulators with enhanced nonsingular fast terminal sliding mode,” IEEE Transactions on Industrial Electronics, vol. 68, pp. 2356–2367, 2021, the control algorithm is executed at a sampling frequency of 1 kHz, and the MAEs of the two controlled joints are 1.30 / 1.26 degrees, respectively. The average MAE of the strategy proposed in this application is 1.92 degrees, executed at a sampling frequency of 100 Hz.

[0119] In summary, the embodiments of the present application propose a new trajectory tracking control scheme for a cable-driven exoskeleton robot with a series elastic actuator. The control objective is achieved in two stages: first, an iterative learning control method is used to approximate the unmodeled disturbance without human participation; second, a suboptimal MPC method is used to drive the robot to track the desired trajectory by compensating for the disturbance and estimating the torque, thereby guiding the human subject to perform the training task. The proposed control scheme has the characteristics of forceless sensors and low complexity, which reduces the requirements for hardware. In fact, it has been successfully applied to high-degree-of-freedom exoskeleton robots with very limited computing resources, and the high-precision results will provide a guarantee for the effectiveness of rehabilitation training in clinical trials.

Claims

1. A trajectory tracking control method for an exoskeleton robot, characterized in that, it includes: In the first stage when the exoskeleton robot is not applied to a subject, determine the system disturbance of the power system of the exoskeleton robot and compensate the output parameters of the power model of the exoskeleton robot; In the second stage when the exoskeleton robot is applied to a subject, isolate the system disturbance from the human - machine interaction torque in the second stage and determine the human - machine interaction torque; Input the human - machine interaction torque into a controller, and use the controller to determine the assistance of the power system of the exoskeleton robot.

2. The method according to claim 1, characterized in that, In the first stage when the exoskeleton robot is not applied to a subject, the steps of determining the system disturbance of the power system of the exoskeleton robot and compensating the output parameters of the power model of the exoskeleton robot include: In the first stage when the exoskeleton robot is not applied to a subject, estimate the system disturbance of the power system of the exoskeleton robot through iterative learning and compensate the output parameters of the power model of the exoskeleton robot.

3. The method according to claim 2, characterized in that, In the step of estimating the system disturbance of the power system of the exoskeleton robot through iterative learning and compensating the output parameters of the power model of the exoskeleton robot, an iterative learning impedance controller following the backstepping method is used to estimate the system disturbance of the power system of the exoskeleton robot.

4. The method according to claim 3, characterized in that, Multiple joints of the exoskeleton robot are all driven by series elastic actuators, and the iterative learning impedance controller is expressed as: The formula (1) represents the dynamic equation of the rigid joint side, and the formula (2) represents the dynamic equation of the series elastic actuator side, where q ∈ R n represents the joint position vector, and θ ∈ R n represents the motor rotor shaft position vector of the joint of the exoskeleton robot, and M(q) ∈ R n×n , and \(g(q)\in\mathbb{R}\) n represent the inertia matrix, centrifugal and Coriolis torques, and gravitational torque of the exoskeleton robot respectively, \(K\in\mathbb{R}\) n×n is the stiffness matrix, \(B\in\mathbb{R}\) n×n is the inertia matrix of the actuator, \(\tau\) e \(\in\mathbb{R}\) n is the physical interaction torque vector, \(\tau\) f \(\in\mathbb{R}\) n represents the disturbance \(u\in\mathbb{R}\) caused by cable transmission and joint system friction of the exoskeleton robot n represents the control torque applied to the series elastic actuator of the exoskeleton robot.

5. The method according to claim 1, characterized in that, In the step of inputting the human - machine interaction torque into a controller and using the controller to determine the assistance of the power system of the exoskeleton robot, the controller is a sub - optimal model predictive controller.

6. The method according to claim 5, characterized in that, The sub - optimal model predictive controller is a sub - optimal model predictive controller following the singular perturbation method.

7. The method according to claim 6, characterized in that, The sub-optimal model predictive controller includes a joint subsystem controller with a relatively slow speed and an actuator subsystem controller with a relatively fast speed. The input of the sub-optimal model predictive controller is expressed as: u Ⅱ = u f + u s where, u f and u s represent the control terms for the fast and slow time scales respectively, and are respectively used to stabilize the joint subsystem and the actuator subsystem.

8. An exoskeleton robot system, including an exoskeleton robot and a series elastic actuator for driving a plurality of active joints of the exoskeleton robot; the exoskeleton robot further includes a processor and a memory, and computer - executable code is stored in the memory. When the computer - executable code is executed, the processor performs the following operations: In the first stage when the exoskeleton robot is not applied to a subject, determine the system disturbance of the power system of the exoskeleton robot and compensate the output parameters of the power model of the exoskeleton robot; In the second stage when the exoskeleton robot is applied to a subject, isolate the system disturbance from the human - machine interaction torque in the second stage and determine the human - machine interaction torque; Input the human - machine interaction torque into a controller, and use the controller to determine the assistance of the power system of the exoskeleton robot.

9. The exoskeleton robot system according to claim 8, characterized in that, The series elastic actuator further includes: a cable, a motor, a first fixing block, a second fixing block, a sleeve, and an elastic element; wherein the tight end of the cable is connected to the sliding member of the motor, the first fixing block can move on the sleeve, the sleeve is fixed on the second fixing block, the loose end of the cable is connected to the joint, when the tight end of the cable moves in the first direction, the first fixing block moves accordingly and compresses the elastic element; when the elastic element is compressed, the second fixing block generates a force towards the first direction; the fixing block moves in the first direction along with the loose end of the cable.

10. The exoskeleton robot system according to claim 9, wherein, each of the series elastic actuators further includes two potentiometers for measuring the compression of the elastic element.

Images