Human exoskeleton cooperative motion multi-stage control method based on gait optimization and fixed time adaptive technology

Abstract

The invention discloses a human exoskeleton cooperative motion multi-stage control method based on gait optimization and a fixed time adaptive technology, is applied to the field of exoskeleton robots, and aims to solve the problem that an existing control method is large in tracking error. According to the invention, a DTW-based GMM-GMR-DMP method is adopted, and a plurality of demonstration tracks are estimated by using GMR. Then, the generated track is used for DMP demonstration, sensor errors and expert errors are reduced to the maximum extent, and meanwhile the generalization ability of the model is reserved; the invention also provides an event triggering method which aims to manage data exchange in a more intelligent way and ensure that updating is carried out when needed, so that the overall performance is improved and limited communication resource challenges are coped with.

Description

Technical Field

[0001] The present invention belongs to the field of exoskeleton robots, and in particular relates to an exoskeleton control technology. Background Art

[0002] Lower-limb exoskeletons have garnered significant attention in recent years due to their potential to assist individuals with limited mobility and augment human capabilities in a variety of applications, such as rehabilitation, industrial support, and military operations. However, effective human-exoskeleton collaborative motion is difficult to achieve due to the inherent nonlinearity of exoskeleton systems. To ensure user comfort, reduce fatigue, and enhance performance, minimizing interaction torque and maintaining compliant control are crucial while addressing the inherent nonlinearity and uncertainty of exoskeleton systems.

[0003] For human-exoskeleton systems, reasonable trajectories are difficult to obtain. Gait planning methods include predefined trajectories, impedance control, and admittance control. However, these methods cannot effectively handle complex motion scenes and are limited in improving wearing comfort. The initial dynamic movement primitives (DMP) are limited to learning one demonstration. If there is interference or noise in the trajectory, the output of the DMP may not be optimal. The Gaussian mixture model and Gaussian mixture regression (GMM-GMR) models have difficulty handling trajectories of different lengths and cannot adjust the reference trajectory frequency. Therefore, considering the advantages of GMM-GMR and DMP, as well as the problem of time difference, it is necessary to propose a method as a high-level layer to estimate the reference trajectory of the intermediate layer.

[0004] Studying the mechanisms of human-exoskeleton interaction control is crucial. Therefore, physical human-robot interaction (PHRI) is achieved by detecting the interaction torque of the human-machine system and then adjusting the corresponding interaction response by modifying the controller parameters. In fact, admittance theory has been widely used in PHRI scenarios. Therefore, admittance control is selected at the intermediate level to obtain the wearer's motion intention. The effectiveness of the admittance controller usually depends on the characteristics of the low-level position controller. Therefore, designing a high-performance position controller such as an intelligent control algorithm is crucial for lower limb exoskeletons. In multiple actual platforms, the tracking error is usually reduced to a small range within a limited time frame. It is necessary to introduce a solution to address the limitations of the necessary information acquisition method.

[0005] In addition, traditional time-triggered control methods require frequent updates and high transmission rates, which often lead to network congestion; there are unnecessary data transmission and calculations, and the disadvantages of low system performance and energy efficiency. Summary of the Invention

[0006] The main purpose of the present invention is to propose a multi-level control strategy based on gait optimization and fixed-time adaptive technology to improve the wearable comfort performance of the operator in human-exoskeleton collaborative motion, aiming to solve the problems and defects existing in the above-mentioned background technology.

[0007] The technical solution adopted by the present invention is: a multi-level control method for human exoskeleton coordinated motion based on gait optimization and fixed time adaptive technology, comprising the following steps:

[0008] S1, generate original gait data by having the tester wear a human exoskeleton;

[0009] S2. Preprocess the original gait data to obtain several original teaching trajectories;

[0010] S3, build a multi-level controller;

[0011] S4, the multi-level controller dynamically adjusts the desired trajectory based on the several original teaching trajectories in step S2 and the human-computer interaction torque obtained in real time, and outputs the actual control torque;

[0012] S5. Based on the dynamically adjusted desired trajectory and actual control torque generated in step S4, the human exoskeleton system is driven to complete the coordinated movement.

[0013] Beneficial effects of the present invention: The present invention adopts a GMM-GMR-DMP method based on dynamic time warping (DTW). The original DMP was limited to learning one demonstration. If there is interference or noise in the trajectory, the output of the DMP may not be optimal. The GMM-GMR model has difficulty processing trajectories of different lengths and cannot adjust the reference trajectory frequency. This method takes into account the advantages of GMM-GMR and DMP, as well as the problem of time difference, and uses GMR to estimate the trajectories of multiple demonstrations. The generated trajectory is then used as a demonstration of the DMP, minimizing sensor and expert errors while retaining the generalization ability of the model.

[0014] Furthermore, traditional time-triggered control methods rely on high update frequencies and transmissions, which often lead to network congestion. To optimize the use of communication resources, this paper proposes an event-triggered approach to reduce update frequency. The Event Triggered Mechanism (ETM) can significantly reduce unnecessary data transmission and computation. In short, ETM aims to manage data exchange in a more intelligent way, ensuring that updates occur only when needed, thereby improving overall performance and addressing the challenges of limited communication resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1This is a schematic diagram of the multi-level control strategy for human-exoskeleton collaborative motion;

[0016] Among them, (a) gait library establishment, (b) high-level gait planning, (c) mid-level admittance control, and (d) low-level position control;

[0017] Figure 2 It is the control block of the exoskeleton system;

[0018] Figure 3 is the general filtered gait experiment result corresponding to the left hip;

[0019] Figure 4 The gait results were obtained using GMM for the left hip;

[0020] Figure 5 is the human-exoskeleton human-machine interaction torque τ ext Schematic diagram;

[0021] Figure 6 is a schematic diagram of the hip joint position response generated by the corresponding position controller in active mode;

[0022] Figure 7 It is the corresponding position controller that causes the tracking errors of the two joints in active mode;

[0023] Figure 8 are the control torques and event triggering time intervals of the two joints in active mode. DETAILED DESCRIPTION

[0024] To facilitate those skilled in the art to understand the technical content of the present invention, the present invention is further explained below with reference to the accompanying drawings.

[0025] See also Figure 1 The present invention provides an exoskeleton control technology with a multi-level control strategy based on gait optimization and fixed-time adaptive technology, in which the high-level control layer is used to learn multiple teaching trajectories and generate reference trajectories suitable for the exoskeleton system, and can adjust the trajectory in real time to identify the operator's movement intention. In addition, due to the use of GMM-GMR-DMP learning technology, the human-exoskeleton interaction torque and position derivative are significantly reduced from the initial stage to the final stage, and the operator's wearing comfort is improved. The admittance controller can switch the passive motion mode to the active motion mode by adjusting parameters. At the low-level layer, a fixed-time adaptive controller based on BLFs is designed to enhance the tracking performance while constraining the tolerable human-exoskeleton interaction torque.

[0026] To implement this technology, a two-degree-of-freedom lower-limb exoskeleton dynamic model with output constraints and joint friction was first established. Secondly, for the design of a multi-level controller, the high-level layer constructed a human gait library, used DTW to align raw demonstration trajectories to a consistent time duration, and established a gait planning method based on GMM-GMR-DMP. Then, the mid-level layer designed an admittance loop to facilitate two different training modes for human-exoskeleton collaborative motion: passive and active. Finally, an adaptive fixed-time control scheme with BLFs was designed at the low-level layer to ensure that joint position errors converge to the zero region within a finite time. Furthermore, an ETM was introduced in the backstepping iteration to reduce communication resources and avoid Zeno behavior. Finally, the proposed method was simulated. Extensive simulation and experimental results demonstrate that the designed control scheme ensures that the two exoskeleton joint angles accurately track the desired trajectories.

[0027] The detailed implementation process of the present invention is as follows:

[0028] S1: Construct a two-degree-of-freedom lower limb exoskeleton dynamic model with output constraints and joint friction. The dynamic equations describe the exoskeleton's motion and force, providing a theoretical basis for the design of a multi-level controller. The expression is:

[0029]

[0030] Where q∈R 2 , and Represent the exoskeleton joint position, velocity and acceleration respectively and are two-dimensional vectors, M(q), and G(q)∈R 2 Represent inertia, Coriolis force and gravity terms respectively. The inertia matrix and Coriolis matrix are both 2×2 matrices, and the weight term is a two-dimensional vector. represents the joint friction term, τ ext represents the interaction torque, τ represents the control torque and both are two-dimensional vectors, R 2 represents a two-dimensional real space. The interaction torque is as follows Figure 5 shown.

[0031] The lower limb exoskeleton in this embodiment can be specifically referred to the patent application with patent application number 202111332323.7. The lower limb exoskeleton system is an existing known technology and will not be introduced in detail in this invention.

[0032] S2: Establish a comprehensive gait database. In the data collection experiment, first calibrate the IMU (Inertial Measurement Unit), plantar pressure sensor and other equipment to ensure data accuracy, and then use a lightweight gait collection device Figure 1 As shown in (a), gait experiments involving various movement patterns are conducted. The lightweight gait acquisition device includes three sensors: SEMG (Surface electromyography) sensor, IMU sensor and plantar pressure sensor. The three sensors are used to synchronously collect three sensor signals. The sample data is obtained from gait experiments in many scenarios, such as Figure 1 As shown in , there are 8 scenes in total, including: stationary, walking on flat ground, climbing stairs, descending stairs, walking uphill, walking downhill, marching in place, squatting and standing. The original gait data of the joints are filtered, as shown in Figure 3 As shown in Figure 1, each trajectory contains three gait cycles with inconsistent time lengths. However, due to some differences in movement speed or walking rhythm, the problem of different gait cycles will be processed using the DTW method in the subsequent steps.

[0033] In this embodiment, low-pass filtering is specifically used to suppress high-frequency interference in the sensor signal and ensure the accuracy of subsequent analysis.

[0034] Figure 3 trj is the abbreviation of trajectory, trj1, trj2, trj3, trj4, and trj5 are five trajectories respectively.

[0035] S3: Design a multi-level controller. The multi-level control strategy for human-exoskeleton collaborative motion consists of three layers, such as Figure 2 As shown, the design can generate reasonable gait trajectories according to the operator's various motion patterns and can ensure the author's wearable comfort by achieving high tracking accuracy of the exoskeleton and minimal relative interaction torque.

[0036] For high-level gait planning, dynamic time warping (DTW) is used to align gait data of different lengths, and Gaussian mixture model (GMM), Gaussian mixture regression (GMR) and dynamic motion primitives (DMP) are used to generate reference trajectories (q r The reference trajectory is used as the input of the intermediate layer admittance controller to adjust the desired trajectory (q d ).

[0037] The DTW algorithm is used to make the teaching trajectory (the original teaching trajectory is passed through Figure 1 (a) is collected from the experiment) and is consistent with different time series characteristics, including changes in data length and rhythm. and The DTW algorithm can be used to perform the best alignment to make their features as consistent as possible, thereby maximizing their similarity. Chinese xis the length of the time series X, that is, the number of time series data points it contains, and each data point has d-dimensional features. Chinese y is the length of the time series Y, that is, the number of time series data points it contains, each of which has d-dimensional features. The DTW algorithm can be described by the following objective function:

[0038]

[0039] Among them, ||·|| F is the F norm, and are two time series signals, representing gait trajectory data of different lengths. x and YW y are two time-ruled curved path matrices used to align the time steps of X and Y, Φ dtw Indicates the aligned signal XW x and YW y The Frobenius norm of is used to measure the difference between the two. Take Φ dtw The minimum value allows for optimal alignment of tracks of different lengths on the time axis.

[0040] GMM is composed of multiple single Gaussian models that smoothly approximate the density distribution of arbitrary shapes. The GMM probability model is used to characterize the teaching trajectory after DTW adjustment and extract common features. The gait results after processing are as follows Figure 4 As shown. For the exoskeleton system of the present invention, the teaching database is described as where y i,s and y i,t Represent the spatial and temporal information of the teaching trajectory respectively. T represents the number of teaching points in the teaching trajectory. For the multidimensional teaching variable y=(y1,y2,...,y T ), GMM can be modeled as

[0041]

[0042] Where p(y) represents the probability density function, K represents the number of Gaussian distributions, and π k represents the weight occupied by the kth Gaussian distribution, N(y; μ k ,Σ k ) represents the Gaussian probability density function, μ k represents the average value, Σ k Denotes the covariance matrix, and D represents the dimension of the teaching trajectory.

[0043] Therefore, the parameters {K,π k ,μ k ,Σ k}. The parameter K can be estimated by Bayesian Information Criterion. The GMM parameter {π k ,μ k ,Σ k}You can use the expectation maximization algorithm to learn in batch mode. This algorithm iteratively optimizes the model by applying the maximum likelihood estimation method. The teaching data of the GMM model is represented as x = [x I ,x O ], where x I represents the query vector, x O represents the encoding vector.

[0044] GMM establishes P(x I ,x O ) probability distribution model is trained, and then GMR is used to calculate the conditional probability P(x I ,x O )'s expectation E(x O |x I ) and covariance COV(x O |x I ), using the calculated expected values as generalized trajectory points to generate a smooth expected trajectory. Using DMP alone to estimate trajectories can introduce unavoidable sensor inaccuracies and expert errors. The combination of GMR and DMP effectively addresses this issue. In this invention, GMR is used to estimate trajectories from multiple examples. The resulting trajectories are then used as examples for DMP, minimizing sensor and expert errors while maintaining the model's generalization capabilities.

[0045] In this paper, a discrete DMP is used to learn the taught trajectory of an individual walking gait and generate a reference trajectory that is used as the input of the admittance control loop. The DMP model ensures that the current system state converges to the desired attractor through the designed nonlinear force term, which is derived from the stable second-order dynamics as shown below

[0046]

[0047] Among them, y, and Represents the position, velocity and acceleration of the exoskeleton joint, which is different from the position, velocity and acceleration q of the exoskeleton joint. and q、 and is the actual value measured by the sensor, where y, and is the virtual state variable of DMP, with slightly different expression symbols, i.e., q is the actual state, y is the ideal trajectory; g represents the desired position of the exoskeleton, corresponding to different motion amplitudes, and η yis the gain coefficient, and F is the nonlinear force term.

[0048] For mid-level admittance control, in order to identify the wearer's movement intention and adjust the gait trajectory in real time through the human-exoskeleton coupling torque, the admittance model of the exoskeleton system is constructed as a mass-spring-damper model:

[0049]

[0050] Where △q=q d -q r ,q d ,q r ∈R 2 is the output and input of the admittance controller, R 2 In mathematics, it represents a two-dimensional real space, where M, B, and K are the inertia, damping, and stiffness matrices. ξ is an adjustable coefficient that determines the exoskeleton's operating mode. ξ = 1 indicates active mode, while ξ = 0 indicates passive mode. If the exoskeleton's human-exoskeleton torque τ is obtained in real time, ext and the reference joint position q r , then the admittance controller output q d The linear transfer function after transformation is calculated as follows:

[0051]

[0052] Among them, the output q d Determined by the admittance parameter and the human-exoskeleton interaction torque, the obtained q d Used for subsequent low-level backstepping technology design.

[0053] For low-level position control, the model uncertainty is first estimated using the fuzzy logic system (FLS): the first IF-THEN rule is represented by R l : If x1 is F1 l ,…,x n yes Then y is G l , l=1,…,N, where F i l and G l is relative to the membership function and The fuzzy set of N is the number of fuzzy rules. The expression of FLS is as follows

[0054]

[0055] in, Define the basis functions so that

[0056]

[0057] Backstepping technology process: The state error of the exoskeleton system can be described as: z = qq d , where z = [z1, z2] T ∈R 2 is the joint tracking error, e=[e1,e2] T ∈R 2 is the velocity error, is the desired demand of the exoskeleton derived from the admittance controller output, β=[β1,β2] T ∈R 2 Denotes a dummy control variable. First, define the following cascade Lyapunov function:

[0058]

[0059] Among them, θ is the boundary value of the output constraint, ensuring that |z i |<θ, M is the inertia matrix of the exoskeleton dynamics model (M(q)∈R 2×2 In this embodiment, the subscript 1 refers to the hip joint, and the subscript 2 refers to the knee joint. For example, z1 represents the tracking error of the hip joint, and z2 represents the tracking error of the knee joint.

[0060] In a physical sense, θ is the safety boundary limit of the joint, which prevents damage to the mechanical structure or discomfort to the user due to excessive errors. In practical applications, constraints are imposed based on the dynamic characteristics of the system.

[0061] Then the derivative of L1 becomes

[0062]

[0063] In addition, β i Can be determined as

[0064]

[0065] Among them, h 1i ,s 1i >0, sgn(.) represents the signum function;

[0066] Substitute the relevant information into the operation, and we can get

[0067]

[0068]

[0069] in, O(X) is the unknown nonlinear dynamics of the system (including friction, unmodeled dynamics, etc.), C, G, and M are the precise model parameters of the exoskeleton. The precise model parameters of the exoskeleton, such as M, C, and G, are difficult to obtain, and there are nonlinear interferences such as friction. Therefore, the unknown parts are collectively represented as O(X) and learned online through a data-driven method (FLS). z=[z1,z2] T , M, C, G, F, α are the inertia matrices (M(q)∈R 2×2 ), Coriolis matrix Gravity term (G(q)∈R 2 ), joint friction and the derivative of the virtual control quantity, α is the derivative of the virtual control quantity β, Used to construct velocity error.

[0070] FLS is used to estimate the uncertainty of the model. The first rule of fuzzy rules (IF-THEN Rules) is represented by R l : If x1 is F1 l ,…,x n yes Then y is G l , l=1,...,N, where F i l and G l is relative to the membership function and The fuzzy set is N, and N is the number of fuzzy rules. The expression of FLS is as follows:

[0071]

[0072] where y l Output fuzzy set G l Typical values,

[0073] In the present invention, FLS is used to estimate the uncertain parameters O i (X i ),as follows:

[0074]

[0075] where Φ i (X i )=[Φ i,1 (X i ),Φ i,2 (X i )] T is the fuzzy basis function vector (another expression of the above formula y(x)),

[0076] is the ideal fuzzy weight vector, which represents the optimal fuzzy rule weight combination, so that the approximation error ε i Minimum. ε i (X i ) is the fuzzy approximation error, which indicates the difference between FLS and the real dynamics O i (X i ) deviation;

[0077] By using Young's inequality, it becomes:

[0078]

[0079] in Represents an unrecognized constant, α i >0;O i O i (X i ) is a shorthand form of .

[0080] Will Formula and e i O i Substitute the inequality into the above formula You can get:

[0081]

[0082] To avoid the Zeno phenomenon of the designed controller, the ETM is established by the following formula

[0083]

[0084] Among them, ω i (t) represents the transient control output, τ i (t) is the actual control output of the designed ETM, y i (t) = ω i (t)-τ i (t) represents the calculation error, 0<ψ i <1,δ i >0,t i,1 represents the initial time, t i,k Indicates the time when the kth event is triggered, k∈Z + , Z + is a positive integer, that is, k = 1, 2,….

[0085] ω i (t) is the event-triggered control torque. When t i,k+1 =inf{t∈R||y i (t)|≥ψ i |τ i(t)|+y1}, the event is triggered, and each event triggers at interval t i,k to t i,k+1 The actual control torque τ i (t) is the event-triggered control torque ω i (t) at t i,k value.

[0086] According to the above ETM formula, the actual control variable τ i Using ETM we can get:

[0087]

[0088] where |γ i,1 (t)|≤1,|γ i,2 (t)|≤1 is a time-varying parameter, ψ represents the event trigger threshold parameter (constant), which is used to control the trigger frequency. i (t) Substitute the actual control variable τ i Combining the Young's inequality with the properties of the hyperbolic tangent function (tanh), we can conclude that:

[0089]

[0090]

[0091] The scaling will e i τ i The complex expression is simplified to a constant upper bound of 0.557δ i , used to prove the derivative of the Lyapunov function Substitute this scaling into the above formula The inequality can be obtained:

[0092]

[0093] Considering FLS weight estimation error The overall Lyapunov function is given by: in r i is a constant, and L3 derivation yields:

[0094]

[0095] Implementation of fixed-time controller: According to the Lyapunov stability law, it must satisfy Negative definite, where the control law The design needs to pass high-order terms and power term -s 2i |e i | 2p-1 sgn(ei ) achieves fixed time convergence, and the unmodeled dynamics are compensated by the fuzzy term offset, middle Item passed design The adaptive law The design should consider eliminating weight estimation errors So the fixed-time convergent controller involves a finite control output Adaptive Law The design is as follows

[0096]

[0097] Among them, h 2i >0,s 2i >0,λ 1i >0,λ 2i represents the gain coefficient of the nonlinear damping term. If the finite control output and adaptive law As given above, all states and estimated errors z i ,e i ,α i All converge to the neighborhood of zero within a fixed time, regardless of the initial state of the system, t k+1,j -t k,j There is a positive lower bound, and Zeno phenomena can be prevented.

[0098] Those skilled in the art should know that the λ 2i Increasing λ can speed up convergence but may cause oscillation. 2i Reducing λ can enhance robustness, but it will slow down convergence; therefore, in this embodiment, 1i and λ 2i Take 1 as the baseline value, and then observe the system response through simulation or experiment in actual application, and gradually adjust λ 2i To balance convergence speed and stability.

[0099] Output of finite control and adaptive law Bring in Through the power sum inequality and Young's inequality, etc., the relevant scaling is performed, and finally all the estimation and state errors z i ,e i ,θ i Both converge and because e i ,z i ,q d , are all bounded, variables q, are also bounded, and ω i,u i is also bounded, therefore, all closed-loop signals are bounded.

[0100] The corresponding position controller generates hip joint position response in active mode, such as Figure 6 The corresponding position tracking error is limited to the specified output range in both active and passive modes, as shown in Figure 7 As shown in Figure 2, the exoskeleton remains compliant and responsive during the interaction with the operator, demonstrating the effectiveness of the proposed conductive control scheme. Therefore, the exoskeleton can follow the exoskeleton's motion intention based on the real-time interaction torque and the operator's active part movement. Furthermore, the maximum event triggering interval is 0.45s, and the minimum event triggering interval is 0.01s. Figure 8 As shown, the Zeno phenomenon associated with the designed controller is effectively prevented. The number of event triggering in both training modes is lower than that of the time-triggered controller, indicating that the proposed controller saves about 50% of communication resources compared to the traditional time-triggered controller.

[0101] Those skilled in the art will appreciate that the embodiments described herein are intended to aid the reader in understanding the principles of the present invention, and it should be understood that the scope of the present invention is not limited to such specific descriptions and embodiments. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims.

Claims

1. A multi-level control method for human exoskeleton coordinated motion based on gait optimization and fixed time adaptive technology, characterized in that: The following steps are involved: S1, generate original gait data by having the tester wear a human exoskeleton; S2. Preprocess the original gait data to obtain several original teaching trajectories; S3, build a multi-level controller; S4, the multi-level controller dynamically adjusts the desired trajectory based on the several original teaching trajectories in step S2 and the human-computer interaction torque obtained in real time, and outputs the actual control torque; S5. Based on the dynamically adjusted desired trajectory and actual control torque generated in step S4, the human exoskeleton system is driven to complete the coordinated movement.

2. The multi-level control method for human exoskeleton coordinated motion based on gait optimization and fixed time adaptive technology according to claim 1, characterized in that: Step S1 synchronously collects three sensor signals, including surface electromyography signals, inertial measurement signals, and plantar pressure signals.

3. The multi-level control method for human exoskeleton coordinated motion based on gait optimization and fixed time adaptive technology according to claim 2, characterized in that: The multi-level controller in step S3 specifically includes: a high-level gait planning unit, a mid-level admittance control unit, and a low-level position control unit; The high-level gait planning unit aligns gait data of different lengths of the original teaching trajectory through dynamic time warping, and generates a reference trajectory using Gaussian mixture model, Gaussian mixture regression and dynamic motion primitives; The intermediate layer admittance control unit is specifically constructed as a second-order admittance model, which outputs a corrected desired trajectory to the low-level position control unit based on the reference trajectory generated by the high-level gait planning unit and the human-computer interaction torque obtained in real time; The low-level position control unit includes a fixed-time controller that estimates the uncertainty of the human exoskeleton online through a fuzzy logic system. Based on the estimated uncertainty of the human exoskeleton, a fixed-time control law is used to generate an event-triggered control torque, and the actual control torque is obtained based on the event-triggered control torque to drive the exoskeleton movement.

4. The multi-level control method for human exoskeleton coordinated motion based on gait optimization and fixed time adaptive technology according to claim 3, characterized in that: The second-order admittance model expression corresponding to the intermediate layer admittance control unit is: Where △q=q d -q r ,q d and q r are the output and input of the admittance controller, M, B and K are the inertia matrix, damping matrix and stiffness matrix, ξ is the adjustable coefficient that determines the working mode of the exoskeleton, τ ext It is the human-computer interaction torque obtained in real time.

5. The multi-level control method for human exoskeleton coordinated motion based on gait optimization and fixed time adaptive technology according to claim 4, characterized in that: ξ=1 is active mode, and ξ=0 is passive mode.

6. The multi-level control method for human exoskeleton coordinated motion based on gait optimization and fixed time adaptive technology according to claim 5, characterized in that: The event-triggered control torque generated based on the fixed-time control law is expressed as: Where ψ represents the event trigger threshold parameter, τ represents the control torque, and τ i (t) is the actual control torque, ω i (t) is the event-triggered control torque generated based on the fixed-time control law, y i (t) = ω i (t)-τ i (t) represents the calculation error, 0<ψ i <1,δ i >0,t i,k Indicates the time when the kth event is triggered, k∈Z + , Z + is a positive integer, and R represents the set of real numbers.

Images