Exoskeleton robot fixed time control method based on output constraint and disturbance observation

Abstract

The invention discloses an exoskeleton robot fixed time control method based on output constraint and disturbance observation, relates to the field of robot control, constructs a fixed time neural controller (FTNC), and ensures that a system is stable in fixed time by combining a universal obstacle Lyapunov function (UBLF) and a preset performance function (PPF), so as to improve the system stability. The convergence time is irrelevant to the initial state; meanwhile, a radial basis function neural network (RBFNN) and a nonlinear disturbance observer (NDO) are adopted to jointly compensate the uncertainty of the model and the man-machine interaction disturbance; by dynamically adjusting UBLF boundary conditions, joint angle errors are limited, and training safety is guaranteed. According to the method, a fixed time control theory is adopted, fixed time is combined with a preset performance function (PPF) and a universal barrier Lyapunov function (UBLF), and the output of the exoskeleton robot is strictly limited, so that the rapid convergence and transient performance requirements of an exoskeleton system are met at the same time.

Description

Technical Field

[0001] The present invention relates to the field of robot control, and in particular to a fixed-time control method for an exoskeleton robot based on output constraints and interference observation. Background Art

[0002] The increasing aging of society and the high incidence of conditions like stroke have led to a yearly increase in the number of patients suffering from lower limb motor dysfunction. Lower limb exoskeleton robots, as a new type of rehabilitation training product, are gradually being applied in the medical rehabilitation field. Compared to traditional methods that rely on the individual experience and skills of rehabilitation therapists and are labor-intensive, rehabilitation robots provide patients with more efficient, targeted, and repeatable training guidance, as well as monitoring and evaluation capabilities.

[0003] Robot-assisted rehabilitation training can be divided into two modes: passive and active. Passive mode uses trajectory tracking control, where the patient's limbs are completely driven by the robot to complete lower limb rehabilitation training. For patients in the early stages of rehabilitation training, whose limb motor abilities have not fully recovered and who cannot perform independent movements, a robot-active, patient-passive rehabilitation training model is essential. This control mode is also widely used in the clinical application of rehabilitation robots.

[0004] In passive mode control, fixed-time convergence (FTC) control methods aim to design a controller that converges the system state to an equilibrium point within a fixed time, independent of the initial state. While the time for finite-time convergence depends on the initial state, the upper bound of FTC time is fixed. The FTC method proposed in Reference 1 (X. Shao, P. Pustina, M. Stölzle et al., “Model-based control for soft robots with system uncertainties and input saturation,” IEEE Transactions on Industrial Electronics, 2023) relies primarily on finite-time stability theory. By designing a high-order sliding mode controller and a corresponding disturbance observer, the method achieves shape control of the soft robot, ensuring convergence within a finite time and independent of the initial conditions.

[0005] During passive control, the force exerted by the human body on the robot can be considered an external disturbance. A nonlinear disturbance observer (NDO) is an effective means of observing external disturbances in real time. A key assumption of the NDO is that the first-order derivative (rate of change) of the external disturbance is zero, which is a simplification. When the human body is used as an external disturbance, the rate of change of the external disturbance is not necessarily zero. Furthermore, without a precise dynamic model, the actual NDO estimation result is a composite of the external disturbance and model uncertainty. The presence of model uncertainty affects the robustness of control. Reference 2 (“System Transformation-Based Neural Control for Full-State-Constrained Pure-Feedback Systems via Disturbance Observer,” IEEE TRANSACTIONS ONCYBERNETICS, Vol. 52, No. 3, March 2022) proposes a neural network-based NDO construction method and verifies the effectiveness of incorporating model uncertainty compensation into the NDO.

[0006] The above literature does not address the passive control problem of exoskeleton robots. The human body is an external disturbance, and its rate of change cannot be simplified to 0, but can be considered bounded. However, the upper bound is unknown, and external disturbance observation and compensation are required. invention content

[0007] In response to the above-mentioned problems of passive control of lower limb exoskeleton robots, the present invention proposes a fixed-time control method for exoskeleton robots based on output constraints and interference observation, and constructs a fixed-time neural network controller (FTNC). By combining the universal obstacle Lyapunov function (UBLF) with the preset performance function (PPF), the system is ensured to be stable within a fixed time, and the convergence time is independent of the initial state. At the same time, the radial basis function neural network (RBFNN) and the nonlinear disturbance observer (NDO) are used to jointly compensate for model uncertainty and human-computer interaction disturbances. By dynamically adjusting the UBLF boundary conditions, the joint angle error is limited to ensure training safety.

[0008] A fixed-time control method for an exoskeleton robot based on output constraints and interference observation is characterized by the following steps:

[0009] Step 1: Establish the dynamic model of the n-DOF exoskeleton robot:

[0010]

[0011] in, is the joint angle, is the joint angular velocity, is the joint angular acceleration; It is the human-machine interaction torque, which acts as an external disturbance during control; is the friction force; is the inertia matrix, which is about function; is the combined matrix of Coriolis force and centripetal force; is the gravity term; The input torque for control.

[0012] Step 2: Construct the state space equation of the system;

[0013] Define the state variable as and , the system input is the control torque , the equation of the state space is established as:

[0014]

[0015] The goal is to Tracking the expected output trajectory , ; The desired state variable is defined as and .

[0016] Step 3: Construct a universal barrier Lyapunov function:

[0017]

[0018] Where, is the barrier Lyapunov function; are variables related to the system state; are system variables that need to be constrained; and To constrain the boundaries, , ;when , that is, the upper and lower boundary values ​​are equal. At this time, the universal barrier Lyapunov function is:

[0019]

[0020] Step 4: Design RBFNN and construct NDO;

[0021] RBFNN is:

[0022]

[0023] Where, ; ; 、 and Represent the input vector, weight matrix and number of neurons respectively; is the output function of the radial basis function neural network;

[0024] NDO is:

[0025]

[0026] in, is the estimated disturbance; is the gain coefficient, and the estimated error is ; is the auxiliary state variable;

[0027] By constructing NDO, the state space equation of the system is rewritten as:

[0028]

[0029] To simplify the equation, define , then the state space equation and NDO expression are simplified to:

[0030]

[0031] .

[0032] Step 5: Based on the universal obstacle Lyapunov function, preset performance function, RBFNN and NDO established in the previous steps, a fixed-time neural controller is constructed. Specifically:

[0033] The trajectory tracking error is defined as:

[0034]

[0035] Where, is the tracking error; is the virtual control error; and is the state variable, is the joint angle, is the joint angular velocity; is the virtual control vector; is the desired state variable.

[0036] A. Used for constraints The barrier Lyapunov function is:

[0037]

[0038] Where, Indicates the number of exoskeleton joints; Denotes the obstacle Lyapunov function of the i-th joint of the exoskeleton, i=1,2,…,n, then:

[0039]

[0040] Where, is a constrained variable ; is the constraint boundary.

[0041] The derivative is:

[0042]

[0043] Finally, we get:

[0044]

[0045] in, is the expected output trajectory of the i-th joint.

[0046] B. Used for constraints The Lyapunov function of is:

[0047]

[0048] Where, ; ; is the deviation between the accurate value and the estimated value of the system inertia matrix.

[0049] then The derivative is:

[0050] .

[0051] At the same time, by designing the control law of NDO ,ensure Less than zero, NDO control law Designed to:

[0052]

[0053] Where, ,and .

[0054] Combined with the NDO control law and related properties, it is finally used to constrain The Lyapunov function is designed as:

[0055] .

[0056] The advantages of the present invention are:

[0057] 1. This invention proposes a fixed-time control method for an exoskeleton robot based on output constraints and interference observation. This method adopts fixed-time control theory and combines fixed time with the preset performance function (PPF) and the universal barrier Lyapunov function (UBLF). This method strictly limits the output of the exoskeleton robot, thereby simultaneously meeting the requirements of rapid convergence and transient performance of the exoskeleton system.

[0058] 2. This paper presents a fixed-time control method for exoskeleton robots based on output constraints and disturbance observation. To address the difficulty in obtaining accurate dynamic models for multi-degree-of-freedom exoskeleton systems, a nonlinear disturbance observer (NDO) based on a radial basis function neural network (RBFNN) is designed. This observer compensates for exoskeleton model uncertainty and external disturbances (such as those caused by the human body) through online estimation, without relying on an accurate inertia matrix.

[0059] 3. This paper proposes a fixed-time control method for an exoskeleton robot based on output constraints and disturbance observation. Based on the system state-space equations, a virtual control variable is designed and a candidate Lyapunov function is constructed. Through appropriate adaptive update laws and parameter design, it is demonstrated that the closed-loop system achieves semi-global practical fixed-time stability (SPFTS) within a fixed time, and all system states remain bounded.

[0060] 4. The fixed-time control method for the exoskeleton robot based on output constraints and interference observation in this invention is verified to be effective through stability analysis.

[0061] 5. The fixed-time control method for an exoskeleton robot based on output constraints and interference observation was experimentally verified using a hip exoskeleton robot with three degrees of freedom. Compared with traditional neural network adaptive control (NAC), the results showed that the fixed-time neural network controller (FTNC) of the present invention can better achieve error convergence and output constraints in actual wearing tests, ensuring the safety of human-computer interaction. The experimental results fully demonstrated the effectiveness of the control method. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flow chart of the fixed-time control method for an exoskeleton robot based on output constraints and interference observation according to the present invention;

[0063] Figure 2 This is a structural block diagram of the fixed-time neural network controller (FTNC) constructed by the fixed-time control method of the exoskeleton robot based on output constraints and interference observation in the present invention;

[0064] Figure 3This is a 3D model diagram of the hip joint rehabilitation exoskeleton robot structure;

[0065] Figure 4 This is a graph showing how weights change over time in the fixed time neural network controller (FTNC) proposed by the present invention;

[0066] Figure 5 is the actual joint angle obtained by using the fixed time neural network controller (FTNC) proposed in this invention;

[0067] Figure 6 is the torque obtained by using the fixed time neural network controller (FTNC) proposed in this invention.

[0068] Figure 7 The following is a comparison chart of the joint angle α tracking error between the fixed time neural network controller (FTNC) and NAC proposed in this invention;

[0069] Figure 8 The figure is a comparison of the joint angle β tracking error between the fixed time neural network controller (FTNC) and NAC proposed in this invention;

[0070] Figure 9 The figure shows the comparison of joint angle γ tracking errors between the fixed time neural network controller (FTNC) and NAC proposed in this invention. DETAILED DESCRIPTION

[0072] The present invention will be described in further detail below with reference to the accompanying drawings.

[0073] The present invention is based on the fixed time control method of the exoskeleton robot with output constraints and interference observation, which is described in detail below. The "." and ".." at the top of the characters represent the first-order derivative and the second-order derivative, "^" represents the estimated value, the upper "-" represents the upper limit of the meaning corresponding to the character, and the lower "-" represents the lower limit of the meaning represented by the character. The system is the controlled wearable exoskeleton system, including its dynamic characteristics, control input and various constraints. The method of the present invention is as follows. Figure 1 、 Figure 2 The specific steps are as follows:

[0074] Step 1: Dynamic modeling;

[0075] like Figure 3 The figure shows a 3D model diagram of the overall robotic exoskeleton for hip joint rehabilitation of the present invention, where q1, q3, and θ2 correspond to the three degrees of freedom of the exoskeleton respectively.

[0076] First, establish the dynamic model of the n-DOF exoskeleton robot:

[0077]

[0078] in, is the joint angle, is the joint angular velocity, is the joint angular acceleration. It is the human-machine interaction torque, which acts as an external disturbance during control; is the friction force; is the inertia matrix, which is about function; is the combined matrix of Coriolis force and centripetal force; is the gravity term; The input torque for control.

[0079] Step 2: Construct the state space equation of the system;

[0080] Define the state variable as and , the system input is the control torque , establish the equation description of the state space:

[0081]

[0082] The goal of the control method of the present invention is to make Tracking the expected output trajectory , ; The desired state variable is defined as and . And the external disturbance of the robot and is bounded, the expected trajectory as well as 、 They are all bounded.

[0083] Step 3: Combine the preset performance function (PPF) and the universal barrier Lyapunov function (UBLF) to constrain the system output and ensure the stability of the system output.

[0084] The universal barrier Lyapunov function (UBLF) is used to ensure that the system output always meets the constraints. The formula is:

[0085]

[0086] Where, is the barrier Lyapunov function (BLF); are variables related to the system state; are system variables that need to be constrained. and To constrain the boundaries, system variables are defined The scope of constraints, , , that is, for all time t, If the constraint boundary is symmetric, that is , then UBLF is:

[0087]

[0088] The definition of UBLF can combine the analysis of constrained and unconstrained systems into a unified structure. In order to better control the dynamic performance, the performance function PPF can be designed as . is the initial value of the upper bound of the constraint, indicating the maximum error range allowed at the beginning of the system. is the final stable value at time t of the upper bound of the constraint, indicating the error range that the system needs to maintain in the long term. k is an adjustable parameter that controls the convergence speed.

[0089] Step 4: Design a radial basis function neural network (RBFNN) and construct a nonlinear disturbance observer (NDO);

[0090] A. Radial Basis Function Neural Network RBFNN is expressed as:

[0091]

[0092] Where, ; ; 、 and Represent the input vector, weight matrix and number of neurons respectively. is the output function of the radial basis function neural network, which takes the input vector Mapped to an output value, which is the weight matrix of the neural network and basis functions It is realized by the linear combination of , which represents the response of the neural network to the input signal. The Gaussian function here is chosen as:

[0093]

[0094] in, For the neuronal centers, It is The width of the Gaussian function of each neuron. In the control method of the present invention, Used to approximate the unknown system function f(Z) by adjusting the weight matrix of the neural network To minimize the approximation error , which enables the neural network to accurately simulate the behavior of the system. Basis function is the basic unit of the neural network. is a vector whose components correspond to the outputs of each neuron in the neural network.

[0095] For large enough , there is an optimal weight matrix in the radial basis function RBFNN neural network and an optimal approximation error , such that:

[0096]

[0097] If the node centers are chosen appropriately, the approximation error is bounded. For analysis purposes only, in a real system the weight matrix To approximate the optimal weight matrix , and assuming that the estimated output Close enough to the true system function f(Z) to ignore the approximation error , then This simplification makes the control design more practical and also reduces the computational complexity.

[0098] B. Preliminary construction of NDO based on state space equations to observe external disturbances.

[0099] When designing NDO, the external disturbance is considered as a constant value, that is, , and thus the NDO is designed as:

[0100]

[0101] in, is the estimated disturbance; is the gain coefficient, and the estimated error is ; is a real matrix with n rows and n columns; is an auxiliary state variable used to construct disturbance estimation; for The first derivative of .

[0102] The design of NDO requires an accurate inertia matrix , the combined matrix of Coriolis force and centripetal force , gravity term Therefore, NDO based on accurate models has defects and deficiencies in practical applications. , , .in, 、 、 are the system inertia matrix, the combined matrix of Coriolis force and centripetal force, and the estimated value of the gravity term, respectively. 、 、 are the deviations between the exact value and the estimated value, respectively.

[0103] Therefore, the state space equation of the system can be rewritten as:

[0104]

[0105] in, is an uncertain term, let it be ; then the state space equation is simplified to:

[0106]

[0107] And simplify the NDO expression to:

[0108]

[0109] Among them, according to the skew symmetry of the exoskeleton dynamic model: Satisfy , ; Satisfy , . represents the Euclidean norm. And:

[0110]

[0111] and represent the minimum and maximum eigenvalues ​​of the matrix, respectively.

[0112] C. Construct a nonlinear disturbance observer (NDO) based on radial basis function neural network (RBFNN).

[0113] Use RBFNN to analyze the overall uncertainty Approximation, we can get:

[0114] .

[0115] Where, is the input vector, which is a comprehensive vector containing and The trajectory tracking error is defined as: Where, is the virtual control vector; is the tracking error; is the virtual control error. is the trajectory tracking error part of. Not directly equivalent to , but a higher-dimensional vector that contains more information about the system state and is used to provide a more comprehensive input to the neural network to better estimate and compensate for the uncertainty in the system. To approximate the error of the neural network, the optimal weight matrix is ​​obtained through iterative update ,make minimized; therefore, The estimated value of can be expressed as: , and , , n is the number of joints.

[0116] make , define RBF estimation error , then the adaptive law of the neural network updating iterative weight matrix is:

[0117]

[0118] Where, is a constant used to control the iteration speed;

[0119] Furthermore, applying Young’s inequality, we can get the By scaling and simplifying the formula, we can get:

[0120]

[0121] Where, Indicates error; It is a design parameter and adjusts the degree of scaling.

[0122] Therefore, the final design of NDO is:

[0123]

[0124] in, , .

[0125] Step 6: Construct a fixed time neural network (FTNC) controller based on the above steps 3 to 5.

[0126] (1) Designed for constraints The UBLF structure is:

[0127]

[0128] Where n represents the number of joints. represents the obstacle Lyapunov function (UBLF) of the i-th joint of the exoskeleton. The constraint corresponding to a specific variable or joint in the system is the total Lyapunov function part of the constraint, ensuring that each constrained variable stability.

[0129]

[0130] The derivative is: ,in .

[0131] Arranging the above formula, we get:

[0132]

[0133] Since it is impossible to directly obtain the exact , so we use the estimated value Perform actual control design. Virtual control vector estimation Designed to:

[0134] in, , , is the control gain of different types of parameters of the i-th joint, k and n correspond to different control gain types

[0135] Type, the numbers in the subscripts correspond to different types of parameters, and the index , .

[0136] In order to avoid too many differential items, The first-order filter is constructed as: ,That

[0137] middle, .

[0138] The virtual control vector estimate Substituting into the formula we get:

[0139] .

[0140] (2) Designed for constraints The Lyapunov function of is:

[0141]

[0142] It is a whole Lyapunov function, which is used to analyze the stability of the whole system and to describe the whole system. The dynamic behavior of May contain multiple variables, each of which corresponds to a joint or state in the system. The design can reflect the relationship between these variables. The formula expression can be obtained .

[0143] right The derivative is:

[0144]

[0145] At the same time, by designing the control law of NDO ,ensure Less than zero, specifically:

[0146] Due to estimation error , using Young's inequality, we can get:

[0147]

[0148]

[0149] Where, is the parameter used to balance the terms on both sides of the inequality. The first inequality uses Young's inequality ,in , .Pick , thus splitting the cross term into two square terms, which facilitates subsequent stability analysis. The second inequality constructs the Lyapunov function: , its derivative Substituting into the error dynamics equation and applying Young's inequality to each term, we can finally get the upper bound of the error dynamics. , which ensures that the coefficient is negative, thus ensuring the stability of the error system, making the estimated error Converges to zero. Therefore, the NDO control law It can be expressed as:

[0150]

[0151] Where, ,and .

[0152] Combining the designed NDO control law and related properties, we can finally obtain the constraint The Lyapunov function of is:

[0153] .

[0154] Step 5: Stability analysis.

[0155] In the exoskeleton robot dynamics model, under the action of the control law and the adaptation law, all signals of the closed-loop system are semi-globally uniformly bounded. According to the form of the first-order filter, the filtering error is , the differential is , .in, It is a concentrated manifestation of uncertainty and disturbance in the system.

[0156] The Lyapunov function is constructed as:

[0157]

[0158] Combining the above formulas, we finally get:

[0159]

[0160] Right now:

[0161]

[0162] The control parameters must meet the following conditions:

[0163]

[0164] So we have:

[0165]

[0166] in,

[0167]

[0168]

[0169] because is bounded and satisfies:

[0170]

[0171] therefore, 、 and are bounded and satisfy , and, ;

[0172] Combining the above formula, we can get:

[0173] So we have:

[0174]

[0175] in,

[0176] #timg# #timg# #timg#

[0177] Therefore, the system is practical finite time stable PFTS, and PPF guarantees transient performance.

[0178] By constructing a composite Lyapunov function and analyzing its derivatives, as well as designing a nonlinear disturbance observer, this paper proves that the system can still achieve finite time stability (PFTS) under disturbance and the transient performance (PPF) is controllable.

[0179] like Figure 4 The figure shows the curve of the weight change over time in the fixed time neural controller (FTNC) proposed by the present invention. The weight convergence and the controller error can be observed. It can be seen in the figure that no matter what the initial error is, The estimated values ​​of can converge to a smaller range within a certain period of time. This shows that the designed adaptive neural network controller can effectively estimate the unknown system dynamics. Even under different initial conditions, the weight estimates can eventually stabilize, providing a stable foundation for the precise control of the robot manipulator.

[0180] Figure 5 and Figure 6 As shown in FIG, the actual joint angle and torque diagrams obtained by using the fixed time neural network controller (FTNC) proposed in this invention are respectively, Figures 7 to 9 Comparisons of the tracking errors for the three joint angles α, β, and γ using the proposed fixed-time neural network controller (FTNC) and the NAC are shown. A significant difference between the two is that the NAC does not use an UBLF for output limitation, but uses an RBFNN for dynamic model compensation, and does not use an NDO for external disturbance observation. The comparison shows that, during wearable experiments, the designed FNTC achieves lower tracking error than the NAC, taking into account external human interference. Furthermore, the time-varying UBLF and timing control ensure that the dynamic tracking error after steady-state operation remains within a certain range. This achieves output constraint and ensures interactive security during passive control, thus validating the effectiveness of the proposed FNTC.

Claims

1. A fixed-time control method for an exoskeleton robot based on output constraints and interference observation, characterized by: The steps are: Step 1: Establish the dynamic model of the n-DOF exoskeleton robot: in, is the joint angle, is the joint angular velocity, is the joint angular acceleration; It is the human-machine interaction torque, which acts as an external disturbance during control; is the friction force; is the inertia matrix; is the combined matrix of Coriolis force and centripetal force; is the gravity term; To control the input torque; Step 2: Construct the state space equation of the system; Define the state variable as and , the system input is the control torque , the equation of the state space is established as: The goal is to Tracking the expected output trajectory , ; The desired state variable is defined as and ; Step 3: Construct a universal barrier Lyapunov function: Where, is the barrier Lyapunov function; are variables related to the system state; are system variables that need to be constrained. and To constrain the boundaries, , ;when , that is, the upper and lower boundary values ​​are equal. At this time, the universal barrier Lyapunov function is: Step 4: Design RBFNN and construct NDO; RBFNN is: Where, ; ; 、 and Represent the input vector, weight matrix and number of neurons respectively; is the output function of the radial basis function neural network; NDO is: in, is the estimated disturbance; is the gain coefficient, and the estimated error is ; is the auxiliary state variable; By constructing NDO, the state space equation of the system is rewritten as: To simplify the equation, define , then the state space equation and NDO expression are simplified to: Step 5: Based on the universal obstacle Lyapunov function, preset performance function, RBFNN and NDO established in the previous steps, a fixed-time neural controller is constructed. Specifically: The trajectory tracking error is defined as: Where, is the tracking error; is the virtual control error; and is the state variable, is the joint angle, is the joint angular velocity; is the virtual control vector; is the desired state variable; (1) Used for constraints The barrier Lyapunov function structure is: Where n represents the number of exoskeleton joints; represents the obstacle Lyapunov function of the i-th joint of the exoskeleton, i=1,2,…,n; then: Where, is a constrained variable ; To constrain the boundaries; The derivative is: ;in , represents the constraint boundary at the initial moment; represents the constraint boundary at time t; Finally, we get: in, is the expected output trajectory of the i-th joint; (2) Used for constraints The Lyapunov function structure is: Where, ; ; is the deviation between the accurate value and the estimated value of the system inertia matrix; then The derivative is: At the same time, by designing the control law u of NDO, we can ensure Less than zero, the NDO control law u is designed as: Where, ,and ; Combined with the NDO control law and related properties, it is finally used to constrain The barrier Lyapunov function structure is designed as: 。 2. A fixed time control method for an exoskeleton robot based on output constraints and interference observation as claimed in claim 1, wherein The characteristic is that in step 2, the Lyapunov preset performance function PPF is designed as: in, is the initial value of the upper bound of the constraint, is the final stable value of the upper bound of the constraint at time t; k is an adjustable parameter that controls the convergence speed.

3. The fixed time control method of an exoskeleton robot based on output constraints and interference observation as claimed in claim 1, wherein The characteristic is that in step 3, there is an optimal weight matrix in RBFNN and an approximation error , such that: In the actual system, by estimating the weight matrix To approximate the optimal weight matrix , and ignore the approximation error , then RBFNN is represented as .

4. The fixed time control method for an exoskeleton robot based on output constraints and interference observation as claimed in claim 1, wherein The characteristic is that in step 3, the state space equation and the simplified NDO expression are: Satisfy , ; Satisfy , ; and: and represent the minimum and maximum eigenvalues ​​of the matrix, respectively.

5. The fixed-time control method for an exoskeleton robot based on output constraints and interference observation as claimed in claim 1, characterized in that: In sub-step B of step 5, Use estimated values Perform actual control design and estimate the virtual control vector Designed to: in, , , is the control gain, and the exponential , .

6. The fixed-time control method for an exoskeleton robot based on output constraints and interference observation as claimed in claim 1, characterized in that: In step six, design for constraints In the process of Lyapunov function, in order to avoid too many differential items, The first-order filter is constructed as: ,in, ; Finally get: 。 7. The fixed-time control method for an exoskeleton robot based on output constraints and interference observation as claimed in claim 1, characterized in that: Using radial basis function neural network RBFNN to estimate ,have to: In the formula, by adjusting the optimal weight matrix , so that the approximation error Minimize; Input vector is a comprehensive vector containing and information; ; is the optimal estimation error; therefore, The estimated value of is: , and ; make ,definition , then the adaptation law for updating the iterative weight matrix is: Where, is a constant used to control the iteration speed; Applying Young's inequality, we can get .

8. The fixed-time control method for an exoskeleton robot based on output constraints and interference observation as claimed in claim 7, characterized in that: NDO is designed to: in, , .

Images