Anti-interference exoskeleton human knee joint torque estimation method based on Kalman filter

Abstract

The present invention discloses a method for estimating the torque of the human knee joint of an anti-interference exoskeleton based on Kalman filtering, which includes the following steps: S1, building a human-machine exoskeleton coupling model based on the OpenSim simulation software to obtain kinematic data; S2, performing Kalman filtering on the kinematic data of the human-machine exoskeleton coupling model; S3, constructing a training set and performing Gaussian process regression fitting for the residual dynamics term of the exoskeleton robot; S4, establishing and solving the dynamic equation of the human knee joint torque; The present invention effectively solves the problem of estimating the human joint torque, reduces the influence of noise, interference, etc., reduces costs, is conducive to the lightweight and portability of exoskeleton technology, and improves the user's comfort.

Description

Technical Field

[0001] The present invention belongs to the field of exoskeleton robots, and more specifically, relates to a method for estimating the torque of an exoskeleton human knee joint based on Kalman filtering and anti-interference. Background Art

[0002] An exoskeleton robot is a human-machine interactive wearable device integrated with multiple technologies such as sensing, control, and mechatronics, and is widely used in scenarios such as medical rehabilitation, walking assistance, and earthquake relief. However, during the human-machine coupled interaction movement, whether it is the walking assistance control system of the exoskeleton or the compliant assistance control that affects the user experience, the motion intention of the human body needs to be known. Currently, the most popular method for estimating the human motion intention is to estimate the joint torque of the human body by combining inverse dynamics and multi-dimensional torque sensors, and then use the human joint torque to replace the human motion intention.

[0003] However, the high cost of multi-dimensional torque sensors, the complexity of the installation and calibration process, and the fitting requirements between the sensors and the human body all limit their popularization in practical applications. At the same time, the use of multi-dimensional torque sensors may affect the structural design of the device, increase the weight and energy consumption of the exoskeleton system, and is not conducive to the development of lightweight and low-cost exoskeleton devices. In addition, as one of the main joints that bear the weight and movement impact of the human body, the knee joint torque estimation faces more challenges. When walking or moving, the knee joint torque is affected by various factors such as inertial force, gravity, and ground reaction force, and the kinematic data such as the actually measured joint angle is often accompanied by noise, which further increases the uncertainty of the dynamic calculation.

[0004] In view of this, the present invention is specifically proposed. By combining dynamic modeling and data-driven methods, it can not only effectively reduce the influence of noise on torque estimation, but also compensate for the residual dynamic terms to improve the estimation accuracy of the knee joint torque. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for estimating the torque of an exoskeleton human knee joint based on Kalman filtering and anti-interference. Without using a force sensor, Kalman filtering is used to filter the noise of the measured kinematic data, and the residual dynamic terms of the exoskeleton are compensated by offline training Gaussian process regression, providing an efficient and low-cost solution for accurate estimation of the knee joint torque.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A method for estimating the torque of an exoskeleton human knee joint based on Kalman filtering and anti-interference includes the following steps:

[0008] S1, building a human-machine exoskeleton coupling model based on OpenSim simulation software to obtain kinematic data;

[0009] S2, perform Kalman filtering on the kinematic data of the human-machine exoskeleton coupling model;

[0010] S3, construct a training set and perform Gaussian process regression fitting on the residual dynamic terms of the exoskeleton robot;

[0011] S4, establish and solve the dynamic equation of the human knee joint torque.

[0012] Preferably, step S1 includes:

[0013] S1.1, construction of the human-machine exoskeleton coupling model: First, use OpenSim simulation software to construct a human lower limb model. The single-leg model includes the thigh, calf, foot contact geometry, hip joint, knee joint, and ankle joint, with a total of nine degrees of freedom; then design the exoskeleton model and add a fixed connection between the exoskeleton model and the human lower limb model to complete the construction of the human-machine exoskeleton coupling model;

[0014] S1.2, simulation of the human movement process: Load gait data and ground reaction force data in the OpenSim simulation software to jointly drive the human-machine exoskeleton coupling model for motion simulation; and run the inverse kinematics tool to obtain the kinematic data of the hip joint and knee joint, including joint angles , angular velocities and angular accelerations ;

[0015] S1.3, storage and export of kinematic data: Save the kinematic data output from the simulation as the data input for the subsequent steps.

[0016] Preferably, step S2 includes:

[0017] S2.1, establish the state transition equation and the observation equation, and discretize to obtain the discrete model: Take the angles and angular velocities of the hip and knee joints of the lower limb exoskeleton human-machine coupling system as state variables, and the joint angular acceleration as the input, and establish the following discrete state equation:

[0018]

[0019] where, , is the hip joint angle, is the hip joint angular velocity, is the knee joint angle, is the knee joint angular velocity; , is the hip joint angular acceleration, is the knee joint angular acceleration; is the discrete time; is the sampling time;

[0020] Observed variables of the lower limb exoskeleton coupling system For the angles and angular velocities of the hip and knee joints, the observation equations are established as follows:

[0021]

[0022] S2.2. According to the state equation and the observation equation, Kalman filtering is used to estimate the joint angles and angular velocities of the hip joint and the knee joint in real time: by adjusting the process noise covariance and the measurement noise covariance , the filtering effect is optimized, and the joint angles and angular velocities obtained after filtering are used as the input for the subsequent steps.

[0023] Preferably, step S3 includes:

[0024] S3.1. Construct a training set: Using the knee joint angle , angular velocity , and angular acceleration as the input, and the residual dynamics term as the output; the residual dynamics term includes friction, noise, and model uncertainty terms; among them, the residual dynamics term can be calculated by the following formula:

[0025]

[0026] Among them, is the inertia matrix of the exoskeleton, is the Coriolis force and centrifugal force matrix, is the gravity matrix, is the joint rotation motor torque (which can be calculated from the current value), is the ground reaction force term;

[0027] S3.2. Gaussian process regression model training: Select the squared exponential kernel function:

[0028]

[0029] Input the kinematic data after Kalman filtering in step S2 into the trained Gaussian process regression model;

[0030] S3.3. Residual dynamics prediction: Use the trained Gaussian process regression model to predict the unmodeled residual dynamics term ; at each moment, combine the kinematic data joint angle , angular velocity and angular acceleration to obtain the corresponding residual dynamics term to achieve the anti-interference effect.

[0031] Preferably, in step S4, the human knee joint torque dynamics equation is as follows:

[0032]

[0033] Where is the inertia matrix of the human-machine exoskeleton coupling model, is the Coriolis force and centrifugal force matrix, is the gravity matrix, is the joint rotation motor torque, is the ground reaction force term, is the predicted residual dynamics term, is the finally estimated human knee joint torque.

[0034] 1. Build a human-machine exoskeleton coupling model through OpenSim software. The simulation data can effectively replace the actually collected motion data, avoiding complex experimental equipment and processes and improving the experimental efficiency;

[0035] 2. By introducing the Kalman filter, the noise of the measurement data is effectively reduced;

[0036] 3. By using Gaussian process regression to fit the residual dynamics term, the anti-interference effect is effectively achieved, and the estimation accuracy of the human joint torque is improved;

[0037] 4. Construct a dynamic estimation model of the human joint torque, providing the core algorithm support for the control strategy optimization of the high-precision torque feedback type exoskeleton robot system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Shows the flowchart of the method of the present invention.

[0039] Figure 2 Shows the human-machine exoskeleton coupling model diagram built through OpenSim software.

[0040] Figure 3 Shows the comparison diagram of the human knee joint torque estimated by the method of the present invention with the true value and the measured value. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] To explain the present invention more clearly, the present invention will be further described below in conjunction with the preferred embodiments and the drawings. Those skilled in the art should understand that the content specifically described below is illustrative rather than restrictive, and should not be used to limit the protection scope of the present invention.

[0042] The anti-interference exoskeleton human knee joint torque estimation method based on the Kalman filter provided in this embodiment is used for human joint torque estimation. As Figure 1 shown, it includes the following steps:

[0043] S1. Based on the OpenSim simulation software, build a human-machine exoskeleton coupling model to obtain kinematic data;

[0044] S2. Perform Kalman filtering on the kinematic data of the human-machine exoskeleton coupling model;

[0045] S3. Construct a training set and perform Gaussian process regression fitting for the residual dynamics term of the exoskeleton robot;

[0046] S4. Establish and solve the dynamic equation of the human knee joint torque.

[0047] Specifically, step S1 includes:

[0048] S1.1. Construction of the human-machine exoskeleton coupling model: First, use the OpenSim software to build a human lower limb model. The single-leg model mainly includes the thigh, calf, foot contact geometry, hip joint, knee joint, and ankle joint, with a total of nine degrees of freedom, which can simulate scenarios such as normal human walking, running, or rehabilitation training; further design the exoskeleton model and add a fixed connection between the exoskeleton model and the human lower limb model to complete the construction of the human-machine exoskeleton coupling model. The model construction diagram is shown in Figure 2 ;

[0049] S1.2. Simulation of the human movement process: Load gait data (such as.trc files) and ground reaction forces (such as.xml files and.mot files) in the OpenSim simulation software to jointly drive the model for motion simulation; and run the inverse kinematics tool (Inverse Kinematics, IK) to obtain the kinematic data of the hip and knee joints, including joint angles , angular velocities and angular accelerations ;

[0050] S1.3. Storage and export of kinematic data: Save the kinematic data (such as.sto files or.mot files) output by the simulation as the data input for the subsequent steps; the simulation data can effectively replace the actually collected motion data, avoiding complex experimental equipment and processes.

[0051] Specifically, step S2 includes:

[0052] S2.1. Establish the state transition equation and the observation equation, and discretize to obtain the discrete model: Take the angles and angular velocities of the hip and knee joints of the lower limb exoskeleton human-machine coupling system as state variables, and the joint angular acceleration as the input, and establish the following discrete state equation:

[0053]

[0054] Among them, , is the hip joint angle, is the hip joint angular velocity, is the knee joint angle, is the knee joint angular velocity; , is the hip joint angular acceleration, is the knee joint angular acceleration; is the discrete time; is the sampling time;

[0055] Observed variables of the lower limb exoskeleton coupling system For the angles and angular velocities of the hip and knee joints, the observation equation is established as follows:

[0056]

[0057] S2.2. According to the state equation and the observation equation, the Kalman filter is used to estimate the joint angles and angular velocities of the hip joint and the knee joint in real time: By adjusting the process noise covariance and the measurement noise covariance , the filtering effect is optimized, and the joint angles and angular velocities obtained after filtering are used as the input for the subsequent steps.

[0058] Specifically, step S3 includes:

[0059] S3.1. Construct a training set: Since the human joint torque cannot be directly measured, the kinematic data when the exoskeleton is unloaded (ensuring that the gait movement trajectory is the same as when loaded) is selected for training, and it is required that the kinematic data cover the entire gait cycle; Using the knee joint angle , angular velocity , and angular acceleration as the input, and the residual dynamics term as the output; The residual dynamics term includes friction, noise, and model uncertainty terms; Among them, the residual dynamics term can be calculated by the following formula:

[0060]

[0061] Among them, is the inertia matrix of the exoskeleton, is the Coriolis force and centrifugal force matrix, is the gravity matrix, is the joint rotation motor torque (which can be calculated from the current value), is the ground reaction force term; All the above information can be measured, so the residual dynamics term can be measured;

[0062] S3.2. Gaussian process regression model training: Select the squared exponential kernel function:

[0063]

[0064] Input the kinematic data after Kalman filtering in step S2 into the trained Gaussian process regression model;

[0065] S3.3, Residual dynamics prediction: Use the trained Gaussian process regression model to predict the unmodeled residual dynamics terms ; At each moment, combine the joint angles of the kinematic data , angular velocities and angular accelerations to obtain the corresponding residual dynamics terms to achieve the anti-interference effect.

[0066] Specifically, in step S4, establish a single-leg human-machine exoskeleton coupled dynamics model, and use the Lagrangian method for dynamics modeling. The human knee joint torque dynamics equation can be obtained as follows:

[0067]

[0068] where, is the inertia matrix of the human-machine exoskeleton coupled model, is the Coriolis force and centrifugal force matrix, is the gravity matrix, is the joint rotation motor torque, is the ground reaction force term, is the predicted residual dynamics term, is the finally estimated human knee joint torque.

[0069] Next, in order to verify the effectiveness of the anti-interference exoskeleton human knee joint torque estimation method based on Kalman filtering provided in this embodiment, MATLAB is used for simulation experiments and verification, and detailed descriptions are given.

[0070] The single-leg human-machine exoskeleton model provided in this embodiment comprehensively considers the influences of noise, friction, model uncertainty, etc., and can accurately estimate the human knee joint torque without using a torque sensor, as follows:

[0071] In the simulation experiment, the thigh mass , the calf mass , the thigh length , the calf length , the distance from the centroid of the thigh to the hip joint , the distance from the centroid of the calf to the knee joint , the moment of inertia of the thigh , the moment of inertia of the calf , the gravitational acceleration , the process noise covariance , measurement noise covariance .

[0072] Based on the above parameters, a simulation experiment is carried out. The human knee joint torque estimated by the method of the present invention is compared with the true value and the measured value. The results are shown in Figure 3 , and it can be seen that the estimation effect of the method of the present invention is good and the filtering effect is remarkable.

[0073] Obviously, the above embodiments of the present invention are only examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or variations can be made based on the above description. It is impossible to list all the implementation manners here. Any obvious changes or variations derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. An anti-interference exoskeleton human knee joint torque estimation method based on Kalman filtering, characterized in that: The steps include: S1, build a human-machine exoskeleton coupling model based on OpenSim simulation software to obtain kinematic data; S2, Kalman filtering is performed on the kinematic data of the human-machine-exoskeleton coupling model; S3, constructing a training set and performing Gaussian process regression fitting on the residual dynamics term of the exoskeleton robot; S4, establish and solve the torque dynamic equation of human knee joint; The step S2 comprises: S2.1, establish the state transfer equation and observation equation, and discretize to obtain the discrete model: take the angles and angular velocities of the hip and knee joints of the human-machine exoskeleton coupling model as state variables, and the joint angular acceleration as input, and establish the following discretized state equation: , in, , is the hip joint angle, is the hip joint angular velocity, is the knee joint angle, is the knee joint angular velocity; , is the hip joint angular acceleration, is the knee joint angular acceleration; It is a discrete moment; is the sampling time; Observable quantities of lower limb exoskeleton coupling system For the angles and angular velocities of the hip and knee joints, the observation equations are established as follows: , S2.2, based on the state equation and observation equation, Kalman filtering is used to estimate the joint angle and angular velocity of the hip and knee joints in real time: by adjusting the process noise covariance and the measurement noise covariance , optimize the filtering effect, and the joint angles and angular velocities obtained after filtering are used as inputs for subsequent steps.

2. The anti-interference exoskeleton human knee joint torque estimation method based on Kalman filtering according to claim 1 is characterized in that: Step S1 includes: S1.1, Construction of human-machine exoskeleton coupling model: First, use OpenSim simulation software to build a human lower limb model. The single-leg model includes the thigh, calf, foot contact geometry, hip joint, knee joint and ankle joint, with a total of nine degrees of freedom; then design the exoskeleton model and add a fixed connection between the exoskeleton model and the human lower limb model, thereby completing the construction of the human-machine exoskeleton coupling model; S1.2, simulate the human motion process: load gait data and ground reaction force data into the OpenSim simulation software, and jointly drive the human-machine exoskeleton coupling model to perform motion simulation; and run the inverse kinematics tool to obtain the kinematic data of the hip and knee joints, including joint angles, angular velocities, and angular accelerations; S1.3, kinematic data storage and export: Save the kinematic data output by the simulation as data input for subsequent steps.

3. The anti-interference exoskeleton human knee joint torque estimation method based on Kalman filtering according to claim 1 is characterized in that: Step S3 includes: S3.1, construct training set: knee joint angle , angular velocity , angular acceleration As input, the residual dynamics term As output; the residual dynamics term includes friction term, measurement noise term and model uncertainty term; among them, the residual dynamics term It is calculated by the following formula: , in, is the inertia matrix of the exoskeleton, is the Coriolis centrifugal force matrix, is the gravity matrix, is the joint rotation motor torque, is the ground reaction force term; S3.2, Gaussian process regression model training: Select the square exponential kernel function: , The kinematic data after Kalman filtering in step S2 is input into the training Gaussian process regression model; S3.3, Residual Dynamics Prediction: Use the trained Gaussian process regression model to predict the unmodeled residual dynamics term ; At each moment, combined with kinematic data joint angle , angular velocity and angular acceleration The corresponding residual dynamics term is obtained to achieve anti-interference effect.

4. The anti-interference exoskeleton human knee joint torque estimation method based on Kalman filtering according to claim 3 is characterized in that: In step S4, the human knee joint torque dynamics equation is: , in, is the inertia matrix of the human-machine-exoskeleton coupling model, is the Coriolis centrifugal force matrix, is the gravity matrix, is the joint rotation motor torque, is the ground reaction force term, is the predicted residual kinetic term, is the final estimated human knee joint torque.

Images