Real-time identification and robust control method for kinetic parameters of lower limb exoskeleton robot

Abstract

The invention relates to the technical field of exoskeleton robots, in particular to a lower limb exoskeleton robot kinetic parameter real-time identification and robust control method, which comprises the steps of S1, system initialization and calibration; s2, motion data acquisition and interference identification; s3, identifying kinetic parameters in real time; s4, target control and compensation calculation; s5, robust control and torque adjustment are carried out; and S6, closed-loop feedback and iterative optimization are carried out. Through a sliding window algorithm with the window length of 0.05 second, the plantar pressure sudden change rate and the joint torque sudden change value are calculated in real time, the sudden interference precursor such as slipping and trip can be accurately captured before motion deviation is really caused by interference, meanwhile, a CAN bus with the transmission rate reaching 1 Mbps is matched, and the joint torque sudden change rate and the joint torque sudden change value are obtained. The delay of transmission of the interference early warning signal from the acquisition link to the compensation link is controlled within 10ms, the interference type can be determined in the initial stage of interference, sufficient time is reserved for subsequent rapid formulation of a coping strategy, and the problem that sudden interference is not perceived in time in a traditional method is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of exoskeleton robot technology, and in particular to a method for real-time identification and robust control of dynamic parameters of a lower limb exoskeleton robot. Background Technology

[0002] The core function of real-time identification and robust control of the lower limb exoskeleton robot's dynamic parameters is to first acquire key dynamic parameters of the robot in real time, including joint inertia, damping, gravity terms, and human-robot interaction forces. Motion and force signals are collected through sensors, and the parameter model is dynamically corrected using adaptive algorithms or extended Kalman filtering to ensure that the parameters match the actual working conditions. Based on this, the robust control section designs control strategies, such as sliding mode control and H∞ control, to address parameter identification errors, external disturbances (e.g., uneven ground, gait changes), and system uncertainties. These strategies maintain the robot's motion stability, accurately track the desired gait trajectory, ensure the wearer's safety and comfort while walking, and enhance the robot's adaptability in different scenarios. For example, when assisting rehabilitation training or walking, the output torque can be adjusted according to the wearer's movements to achieve human-robot coordinated movement.

[0003] However, during the implementation of the above technical solution, at least the following technical problems were discovered:

[0004] Current exoskeletons primarily identify interference by collecting historical data, such as pressure and angle changes over the past 0.05 seconds, and using threshold judgments. For example, a sudden pressure change exceeding 100N is considered slippage. However, this post-analysis delay of approximately 0.03-0.05 seconds exceeds the duration of the interference (0.05-0.1 seconds), causing the interference to cause deviations during identification and making it impossible to anticipate and respond in advance. In contrast, when the human body encounters interference, such as an impending slippage, it can anticipate it 0.1-0.2 seconds in advance through foot tactile and muscle perception. Exoskeletons lack similar active sensing sensors, such as foot tactile sensors and electromyography (EMG) sensors, and cannot simulate the anticipatory abilities of the human body. Existing robust control compensation schemes cannot pre-determine the correspondence between all interference scenarios and compensation torques. As a result, when interference occurs, robust control can only execute the original strategy due to the lack of a pre-defined scheme, and cannot flexibly adjust the compensation force and direction, thus posing a risk of imbalance to the wearer. Summary of the Invention

[0005] Technical problem to be solved: Current exoskeletons mainly identify interference by "collecting historical data, such as pressure and angle changes in the past 0.05 seconds + threshold judgment". For example, a sudden pressure change exceeding 100N is judged as slippage. However, this post-analysis delay of about 0.03-0.05 seconds exceeds the duration of the interference of 0.05-0.1 seconds. As a result, the interference has already caused deviation by the time of recognition, making it impossible to deal with it in advance. When the human body encounters interference, such as when it is about to slip, it can predict it 0.1-0.2 seconds in advance through foot tactile and muscle perception. However, exoskeletons lack similar active perception sensors, such as foot tactile sensors and electromyography sensors, and cannot simulate the predictive ability of the human body. Existing robust control compensation schemes cannot preset the correspondence between all interference scenarios and compensation torques. As a result, when interference occurs, robust control can only execute the original strategy because there is no preset scheme. It cannot flexibly adjust the compensation force and direction, which will cause the wearer to lose balance.

[0006] To address the shortcomings of existing technologies, this invention provides a method for real-time identification and robust control of dynamic parameters of a lower limb exoskeleton robot, thereby solving the technical problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for real-time identification and robust control of dynamic parameters of a lower limb exoskeleton robot includes the following steps:

[0009] S1: System initialization and calibration. Start the joint motors, sensor modules and data transmission modules of the lower limb exoskeleton. Confirm that the hardware is fault-free and the communication is normal. Input static parameters such as the wearer's height, weight and lower limb length as the initial prior values ​​of the dynamic model. Perform zero-point calibration on the encoder and force sensor. Place the joint in the extension zero position to calibrate the angle reading. Zero the force sensor under no-load conditions.

[0010] S2: Motion data acquisition and interference identification. It acquires joint angle, angular velocity, motor drive torque and plantar pressure data at a frequency of 300Hz. High-frequency noise is eliminated by Kalman filtering. A sliding window algorithm is embedded to calculate the plantar pressure mutation rate and joint torque mutation value, and to generate interference warning signals in real time.

[0011] S3: Real-time identification of dynamic parameters. Based on the exoskeleton two-link dynamic model, the collected angle and torque data are used as input. The least squares method is used for slow dynamic motion and the Golden Jackal algorithm is used for fast dynamic motion. The inertia matrix, Coriolis force coefficient, gravity coefficient and joint damping coefficient are solved to output the accurate dynamic parameters under the current motion state.

[0012] S4: Control target and compensation calculation. Generate joint target trajectory based on the wearer's movement intention. In the flat walking scenario, the hip joint extends 10° in the support phase and flexes 30° in the swing phase. The knee joint flexes 5° in the support phase and 60° in the swing phase. Analyze the model estimation deviation, transmission mechanism friction and other common interferences. Combined with the interference type in S2, calculate the compensation torque through a linear formula to obtain the exclusive compensation torque.

[0013] S5: Robust control and torque adjustment. The Udwadia-Kalaba robust control strategy is adopted. The nominal torque to ensure trajectory tracking and the robust compensation torque to offset conventional disturbances are calculated. The two are superimposed with the exclusive compensation torque of S4 to obtain the total control torque. The actual output deviation is verified by the torque sensor before the control is issued.

[0014] S6: Closed-loop feedback and iterative optimization, collects the actual joint angle and angular velocity after the motor is executed, compares them with the target trajectory to calculate the angle error, and immediately corrects the subsequent target trajectory after interference occurs.

[0015] In one possible implementation, in step S2, the sliding window length is set to 0.05 seconds;

[0016] In the embedded sliding window algorithm, the plantar pressure mutation rate is set to R. F ;

[0017] Plantar pressure mutation rate F(t) is the plantar pressure at the current moment, and F(t-0.01) is the plantar pressure 0.01 seconds ago. If (F(t)>2000N / s), it is determined to be a sign of impending slippage.

[0018] In one possible implementation, in step S2, the sudden change value of the joint torque is set as Δτ, which is used to identify the signs of a tripping incident.

[0019] The sudden change value of joint torque Δτ = |τ(t) - τ(t-0.01)|, where τ(t) is the joint torque before the moment of impact and τ(t-0.01) is the joint torque 0.01 seconds before the moment of impact. If (Δτ>5N·m), it is determined to be a "tripping precursor".

[0020] In one possible implementation, in step S4, the magnitude of the compensation torque is dynamically determined by a linear calculation formula;

[0021] If slippage is determined, then calculate the slippage compensation torque τ. comp1 τ comp1 =K F ×R F K F The coefficients are experimentally calibrated, and K F =0.0015, which can convert the pressure mutation rate into the corresponding braking torque.

[0022] In one possible implementation, in step S4, if a trip is determined, then the trip compensation torque τ is calculated. comp2 τ comp2 =K τ ×Δτ, where K τ The coefficients are experimentally calibrated, and K τ =0.8, which can convert the sudden torque value into the corresponding unloading and buffer torque.

[0023] In one possible implementation, in step S5, the torque formula of the Udwadia-Kalaba robust control strategy is transformed into: τ total =τ0+τ r +τ comp Where τ0 is the nominal torque, τ r It is a traditional robust compensation torque.

[0024] In one possible implementation, in step S5, when τ is issued... total Previously, the actual output torque τ was collected using a joint torque sensor. actual Then calculate the deviation Δτ check =|τ total =τ actual |;

[0025] If Δτ check If the torque output is less than 1 N·m, and motor drive friction causes insufficient torque output, then immediately fine-tune τ. total .

[0026] In one possible implementation, the parameter settings of the Golden Jackal algorithm in step S3 include: population size of 30, maximum number of iterations of 500, initial exploration factor of 2.0, and initial development factor of 0.5.

[0027] In one possible implementation, in step S6, the control period for the closed-loop feedback and iterative optimization is 10 milliseconds.

[0028] Beneficial effects compared to existing technologies:

[0029] 1. In terms of interference identification, this solution uses a sliding window algorithm with a window length of 0.05 seconds to calculate the sudden change rate of plantar pressure and the sudden change value of joint torque in real time. It can accurately capture sudden interference precursors such as slipping and tripping before the interference actually causes movement deviation. At the same time, with the CAN bus with a transmission rate of 1Mbps, the delay of transmitting the interference warning signal from the acquisition stage to the compensation stage is controlled within 10ms. The interference type can be identified at the initial stage of interference, leaving sufficient time for subsequent rapid formulation of response strategies. This effectively solves the problem that sudden interference is not detected in time in traditional methods.

[0030] 2. In terms of compensation torque calculation, this scheme uses a linear formula to dynamically determine the magnitude of the compensation torque based on different types of disturbances. The slippage compensation coefficient of 0.0015 was obtained by fitting 100 sets of ground experiments with different friction coefficients, and the tripping compensation coefficient of 0.8 was verified and adjusted through multi-height obstacle scenarios. It can generate customized compensation torques for various sudden disturbances, and the linear calculation logic is simple, taking less than 0.01 seconds. It ensures both adaptability and real-time performance, successfully solving the problem of robust control without a preset corresponding compensation scheme.

[0031] 3. In terms of robust control execution, this solution improves the torque formula of the Udwadia-Kalaba robust control strategy by superimposing the nominal torque, the traditional robust compensation torque, and the special compensation torque for sudden disturbances. At the same time, it combines a torque verification mechanism based on PI control logic. When the deviation between the actual output and the expected torque exceeds 1 N·m, the motor current is automatically fine-tuned without changing the control strategy. This can quickly correct the deviation and ensure accurate torque execution, effectively avoiding control instability caused by sudden disturbances.

[0032] 4. In terms of dynamic parameter identification, this scheme adopts differentiated algorithms for slow and fast dynamic motions. For slow dynamic motions, such as walking on flat ground, the least squares method is used to ensure the accuracy of parameter fitting. For fast dynamic motions, such as running, the Golden Jackal algorithm with specific parameters is used, with a population size of 30 and a maximum number of iterations of 500. The exploration and development factors are dynamically adjusted, which can output accurate dynamic parameters that fit the current motion state, avoiding the problem of insufficient accuracy or slow convergence of a single algorithm, and providing a reliable model basis for robust control.

[0033] 5. In terms of closed-loop feedback optimization, this solution uses a control cycle of 10 milliseconds. Each joint movement cycle is approximately 200ms, which can complete 20 feedback adjustments. When the angle error exceeds 3°, the sensor is recalibrated by standard force and angle input, and the number of iterations of the identification algorithm is optimized. After interference occurs, the target angle of the joint is corrected by 2°-5° within the next 0.1 seconds, while the movement speed is reduced by 10%-20%, providing a buffer for the system to recover stability and further improving the ability to cope with interference. Attached Figure Description

[0034] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0035] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0036] Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can also be implemented in various different forms, and therefore the present invention is not limited to the embodiments described below.

[0037] The technical solution in this application embodiment is to solve the problems mentioned in the background art, and the overall idea is as follows:

[0038] Example:

[0039] Please refer to Figure 1 As shown in the figure, this embodiment introduces a method for real-time identification and robust control of dynamic parameters of a lower limb exoskeleton robot, including the following steps:

[0040] S1: System initialization and calibration. Start the joint motors, sensor modules and data transmission modules of the lower limb exoskeleton. Confirm that the hardware is fault-free and the communication is normal. Input static parameters such as the wearer's height, weight and lower limb length as the initial a priori values ​​of the dynamic model. Perform zero-point calibration on the encoder and force sensor. Place the joint in the extension zero position to calibrate the angle reading. Zero the force sensor under no-load condition to ensure that there are no systematic errors in subsequent data acquisition.

[0041] S2: Motion data acquisition and interference identification. It acquires joint angle, angular velocity, motor drive torque, and plantar pressure data at a frequency of 300Hz. High-frequency noise is eliminated by Kalman filtering. A sliding window algorithm is embedded to calculate the plantar pressure mutation rate and joint torque mutation value. When the pressure mutation rate exceeds 2000N / s, it is determined to be slipping. When the torque mutation value exceeds 5N·m, it is determined to be tripping. Interference warning signals are generated in real time.

[0042] S3: Real-time identification of dynamic parameters. Based on the two-link dynamic model of the exoskeleton, the collected angle and torque data are used as input. The least squares method is used for slow dynamic motion and the Golden Wolf algorithm is used for fast dynamic motion. The inertia matrix, Coriolis force coefficient, gravity coefficient and joint damping coefficient are solved to output the accurate dynamic parameters under the current motion state, providing a model basis for robust control.

[0043] S4: Control target and compensation calculation. Generate joint target trajectory based on the wearer's movement intention. In the flat walking scenario, the hip joint extends 10° in the support phase and flexes 30° in the swing phase. The knee joint flexes 5° in the support phase and 60° in the swing phase. Analyze the model estimation deviation, transmission mechanism friction and other common disturbances. Combined with the disturbance type in S2, calculate the compensation torque through a linear formula. When slipping, multiply the pressure mutation rate by a coefficient of 0.0015. When tripping, multiply the torque mutation value by a coefficient of 0.8 to obtain the exclusive compensation torque.

[0044] S5: Robust control and torque adjustment. The Udwadia-Kalaba robust control strategy is adopted to calculate the nominal torque to ensure trajectory tracking and the robust compensation torque to offset conventional disturbances. The two are superimposed with the exclusive compensation torque of S4 to obtain the total control torque. Before the output is sent, the actual output deviation is checked by the torque sensor. When the deviation exceeds 1 N·m, the motor drive current is adjusted to ensure accurate execution of the total torque.

[0045] S6: Closed-loop feedback and iterative optimization. Collect the actual joint angle and angular velocity after the motor is executed, compare them with the target trajectory to calculate the angle error. When the error is less than 1°, maintain the current identification algorithm and control parameters. When the error exceeds 3°, recalibrate the sensor, optimize the number of iterations of the identification algorithm, or adjust the robust compensation coefficient. After the interference occurs, immediately correct the subsequent target trajectory and reduce the movement speed to offset the interference effect.

[0046] In step S2, the sliding window length is set to 0.05 seconds and embedded in the sliding window algorithm, and the plantar pressure mutation rate is set to R. F Plantar pressure abrupt change rate is used to identify early signs of slippage. During slippage, the friction between the foot and the ground drops sharply, and the pressure fluctuates significantly within 0.01 seconds. Plantar pressure abrupt change rate... F(t) is the plantar pressure at the current moment (unit: N), and F(t-0.01) is the plantar pressure 0.01 seconds ago. If (F(t)>2000N / s) (the threshold calibrated through experiments, corresponding to the rate of pressure change during slippage), it is determined to be a precursor to slippage. The implementation process of the sliding window algorithm is as follows: with a fixed time window of 0.05 seconds, the real-time collected plantar pressure data is rolled and intercepted. The data in the window is updated every 0.01 seconds to ensure that the instantaneous characteristics of pressure change can be continuously captured and to avoid missing the precursor to interference due to untimely data interception.

[0047] In step S2, the sudden change value of joint torque is set as Δτ. Δτ is used to identify the precursor to tripping. When tripping over an obstacle, the joint will be subjected to an instantaneous impact, and the torque will suddenly change. The sudden change value of joint torque Δτ = |τ(t) - τ(t-0.01)|, where τ(t) is the joint torque before the tripping (unit: N·m), and τ(t-0.01) is the joint torque 0.01 seconds ago. If (Δτ>5N·m) (the impact threshold calibrated in the experiment), it is determined to be a precursor to tripping. This solves the pain point of not recognizing sudden interference in real time and avoids the problem of not being able to distinguish between normal fluctuations and precursors of interference when collecting regular motion data. By quantifying the pressure mutation rate and torque mutation value, the method can capture signals before the interference actually causes motion deviation (such as when slippage has just begun and before the lower leg has slipped), generate interference warnings (labeled as interference type: slippage / tripping), and synchronously transmit the warning signal to S4 (control target and interference analysis) to provide a basis for subsequent compensation. To ensure the timeliness of the interference warning signal transmission, the CAN bus is used as the core communication protocol of the data transmission module, with a transmission rate of up to 1Mbps. It can control the delay of transmitting the interference warning signal from S2 to S4 within 10ms, meeting the time requirements of real-time compensation.

[0048] In step S4, the magnitude of the compensation torque is dynamically determined through a linear calculation formula to avoid the problem of poor adaptability of fixed compensation force. If slippage is determined, the slippage compensation torque τ is calculated. comp1 τ comp1 =K F ×R F K F The coefficients calibrated experimentally (unit: N·m·s / N), and K F =0.0015, which can be used to convert the pressure mutation rate into the corresponding braking torque. For example, when R F When τ = 2000 N / s, comp1 =0.0015×2000=3N·m. This torque is just enough to suppress the forward swing of the lower leg caused by slippage, and will not cause joint jamming due to excessive torque. The calibration process of this coefficient 0.0015 is as follows: 100 sets of slippage experiments were carried out on simulated ground with different friction coefficients (0.1-0.6). The relationship between the pressure mutation rate and the required braking torque in each set of experiments was recorded. The result was obtained by fitting through linear regression analysis to ensure that an appropriate compensation torque can be provided under different degrees of slippage.

[0049] In step S4, if it is determined that the person has tripped, then the tripping compensation torque τ is calculated. comp2 τ comp2 =K τ ×Δτ, where K τ The coefficients are experimentally calibrated, and K τ=0.8, which can convert the sudden torque value into the corresponding unloading and buffering torque. For example, when Δτ = 5 N·m (typical tripping impact), τ comp2 =0.8×5=4N·m. This torque can quickly reduce the joint driving force, avoid leg stiffness caused by impact, solve the problem of robust control without preset compensation scheme, and avoid the phenomenon that only analyzing normal disturbances cannot deal with sudden situations such as slipping and tripping. The newly added calculation formula can generate a customized compensation torque for each sudden disturbance through the dynamic correlation of feature value and compensation torque. The calculation logic is simple (linear formula) and takes less than 0.01 seconds, meeting the real-time requirements. At the same time, the compensation torque is passed to S5 to provide a basis for control execution. The calibration of the coefficient 0.8 is determined by setting obstacles of different heights (1-5cm) in a laboratory environment, simulating 100 tripping scenarios, measuring the relationship between the sudden change value of joint torque and the required unloading torque in each scenario, and determining it after multiple verifications and adjustments to ensure that it can effectively buffer tripping impacts of different intensities.

[0050] In step S5, the torque formula of the Udwadia-Kalaba robust control strategy is transformed into: τ total =τ0+τ r +τ comp Where τ0 is the nominal torque, the basic torque that ensures normal trajectory tracking, τ r It is a traditional robust compensation torque, τ comp Specifically designed to counteract sudden disturbances such as slippage and tripping, and used to counteract conventional disturbances such as model errors and minor friction, the three combined achieve a dual effect of conventional stability and sudden disturbance immunity. If τ0=5N·m, τ r = 1 N·m (to compensate for conventional errors), S4 input τ comp =3 N·m, then the modified total torque τ total =5 + 1 + 3 = 9 N·m. At this point, the hip joint actually outputs a braking torque of 9 N·m, which can quickly suppress the forward swing of the lower leg caused by slippage and avoid the angle deviation from increasing (from the original possible 10° to <2°). If it is a tripping scenario τ comp2 = 4 N·m (unloading compensation), then τ total =5+1-4=2N·m (the negative sign indicates that the torque direction is reversed, that is, the driving force is reduced), which quickly buffers the impact and avoids joint stiffness.

[0051] In step S5, when issuing τ total Previously, the actual output torque τ was collected using a joint torque sensor. actual Then calculate the deviation Δτ check =|τ total =τ actual |, if Δτ check If the torque output is less than 1 N·m, and motor drive friction causes insufficient torque output, then immediately fine-tune τ.total (For example, increasing the motor drive current by 5%) ensures accurate execution of the compensation torque, avoids interference response failure due to execution deviation, solves the problem of subsequent control instability, and avoids the phenomenon that the original control law can only cope with normal interference, and the torque is insufficient when encountering sudden interference, resulting in the expansion of trajectory deviation. The modified formula adds τ comp It can provide additional torque support for sudden disturbances, quickly correct deviations, and does not require changing the control strategy (reusing the stability of the original robust strategy). At the same time, the auxiliary verification logic ensures accurate torque execution and ultimately avoids control instability. The fine adjustment of the motor drive current adopts proportional-integral (PI) control logic. With the torque deviation as input, the required adjustment current value is calculated through the preset proportional coefficient (0.2A / (N·m)) and integral coefficient (0.05A / (N·m·s)), ensuring smooth and rapid convergence of current adjustment and avoiding motor vibration caused by sudden current changes.

[0052] In step S3, the parameters of the Golden Jackal algorithm are set as follows: population size is 30, maximum number of iterations is 500, initial exploration factor is 2.0, and initial development factor is 0.5. During the algorithm iteration process, the exploration factor will decrease linearly from 2.0 to 0.5 as the number of iterations increases, while the development factor will increase linearly from 0.5 to 2.0. This balances the global search capability in the early stage of the algorithm with the local optimization accuracy in the later stage, ensuring that the optimal dynamic parameters can be found quickly in fast dynamic motion (such as running and jumping).

[0053] In step S6, the control cycle of the closed-loop feedback and iterative optimization is 10 milliseconds. The control cycle is set based on the fact that the highest frequency of lower limb exoskeleton joint movement is about 5Hz (corresponding to a cycle of 200ms). A control cycle of 10 milliseconds can complete 20 feedback adjustments within each joint movement cycle, ensuring timely correction of trajectory deviations, while avoiding excessive computational load on the system due to an excessively short control cycle.

[0054] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for real-time identification and robust control of lower extremity exoskeleton robot dynamics parameters, comprising the following steps: S1: system initialization and calibration, starting the joint motor, sensor module and data transmission module of the lower extremity exoskeleton, confirming that the hardware is fault-free and the communication is normal, inputting the static parameters such as the height, weight and lower limb length of the wearer as the initial prior value of the dynamics model, zero-point calibrating the encoder and force sensor, placing the joint at the straight zero position to calibrate the angle reading, and clearing the force sensor in the no-load state; S2: motion data acquisition and disturbance identification, collecting joint angle, angular velocity, motor driving torque and plantar pressure data at a frequency of 300 Hz, eliminating high-frequency noise through Kalman filtering, embedding a sliding window algorithm, calculating the plantar pressure mutation rate and joint torque mutation value, and generating a real-time disturbance warning signal; S3: real-time identification of dynamics parameters, based on the two-link dynamics model of the exoskeleton, using the collected angle and torque data as input, using least squares method for slow dynamic motion and using golden wolf algorithm for fast dynamic motion to solve the inertia matrix, Coriolis force term coefficient, gravity term coefficient and joint damping coefficient, and outputting accurate dynamics parameters under the current motion state; S4: control target and compensation calculation, generating joint target trajectory according to the wearer's motion intention, extending 10° in the hip joint support phase and flexing 30° in the swing phase on flat ground, flexing 5° in the knee joint support phase and flexing 60° in the swing phase, analyzing model estimation deviation, transmission mechanism friction and other regular disturbances, combining the disturbance type of S2, calculating the compensation torque through a linear formula, and obtaining the exclusive compensation torque; S5: robust control and torque adjustment, using Udwadia-Kalaba robust control strategy to calculate the nominal torque for ensuring trajectory tracking and the robust compensation torque for offsetting regular disturbances, superimposing the two with the exclusive compensation torque of S4 to obtain the total control torque, and checking the actual output deviation through the torque sensor before issuing; S6: closed-loop feedback and iterative optimization, collecting the actual joint angle and angular velocity after the motor executes, comparing with the target trajectory to calculate the angle error, and immediately correcting the subsequent target trajectory after the disturbance occurs.

2. The method of claim 1, wherein the method is a method of real-time identification and robust control of dynamics parameters of a lower extremity exoskeleton robot, characterized in that, In the step S2, the sliding window length is set to 0.05 seconds; The rate of plantar pressure mutation is set as R in the embedded sliding window algorithm F ; Wherein, the foot pressure mutation rate F(t) is the current foot pressure, F(t-0.01) is the foot pressure 0.01 seconds ago, and if (F(t)>2000N / s), it is determined as a slip precursor.

3. The method of claim 1, wherein the method is characterized by: In the step S2, the joint torque mutation value is set to Δτ, and Δτ is used to identify the tripping precursor; Wherein, the joint torque mutation value Δτ = |τ(t)-τ(t-0.01)|, τ(t) is the front joint torque, τ(t-0.01) is the joint torque 0.01 seconds ago, and if (Δτ>5N·m), it is determined as a tripping precursor.

4. The method of claim 2, wherein the method is characterized by, In the step S4, the compensation torque size is dynamically determined through a linear calculation formula; If it is determined that the wheels are slipping, a slip compensation torque τ comp1 : τ comp1 = K F × R F , where K F is a coefficient experimentally calibrated, and K F = 0.0015, the rate of pressure jump can be converted into the corresponding braking torque.

5. The method of claim 3, wherein the method is characterized by, In the step S4, if it is determined that a stumble has occurred, a stumble compensation torque τ comp2 : τ comp2 τ = K τ × Δτ, where K τ is a coefficient of experimental calibration, and K τ = 0.8, the torque jump value can be converted into a corresponding unloading and cushioning torque.

6. The method of claim 1, wherein, In the step S5, the torque formula of the Udwadia-Kalaba robust control strategy is changed to: total = τ0+ τ r + τ comp , where τ0is the nominal torque, τ r is the traditional robust compensation torque.

7. The method of claim 6, wherein the method is a method of real-time identification and robust control of the dynamics of a lower extremity exoskeleton robot, characterized in that, The step S5, before issuing τ total , the actual output torque τ actual is collected by the joint torque sensor check , the deviation Δτ total is calculated again actual ; If Δτ check <1N·m, if the motor drive friction causes the torque output to be insufficient, immediately fine-tune τ total .

8. The method of claim 1, wherein, In the step S3, the parameter settings of the golden wolf algorithm include: population size is 30, maximum iteration number is 500, initial exploration factor is 2.0, and initial development factor is 0.

5.

9. The method of claim 1, wherein, In the step S6, the control cycle of the closed-loop feedback and iterative optimization is 10 milliseconds.

Images