Human-in-the-loop adaptive control method and system for walking-assist exoskeleton based on energy consumption optimization

Abstract

The present invention discloses a human-in-the-loop adaptive control method for a walking-assist exoskeleton based on energy consumption optimization, comprising the following steps: clustering historical control parameters of the exoskeleton to obtain several groups of typical control parameters, collecting the instantaneous metabolic rate of a subject under several groups of the typical control parameters, and fitting the instantaneous metabolic rate to obtain a steady-state metabolic rate; inputting the several groups of the typical control parameters and the corresponding steady-state metabolic rate into a Bayesian optimization model, the Bayesian optimization model outputting a control parameter combination that minimizes the steady-state metabolic rate as the current optimal control parameter; obtaining the subject's motion trajectory under the optimal control parameter, fitting the motion trajectory to obtain the subject's expected trajectory, driving the exoskeleton according to the expected trajectory, and updating the input of the Bayesian optimization model according to the optimal control parameter and the subject's steady-state metabolic rate at this time; and repeating the above steps until the subject completes the rehabilitation exercise.

Description

Technical Field

[0001] The present invention relates to the field of rehabilitation medical technology, and in particular to a method and system for adaptively controlling a walking-assisting exoskeleton human-in-the-loop based on energy consumption optimization. Background Art

[0002] Wearable exoskeletons are a rehabilitation technology designed to assist patients in walking while reducing their energy expenditure. Despite their significant potential, practical applications face numerous challenges and uncertainties in adjusting exoskeleton parameters to achieve optimal performance. To address these challenges, a human-in-the-loop optimization strategy was introduced. This method evaluates and adjusts parameters in real time based on the subject's energy expenditure to find the optimal configuration with the lowest energy consumption.

[0003] Human-in-the-loop optimization strategies have been applied in various fields and demonstrated promising control results. For example, in autonomous driving, effective remote control technology has been achieved by integrating human interaction feedback into algorithms, collecting driver instructions for data training, and operating unmanned vehicles in virtual scenarios. In medical rehabilitation, human-in-the-loop control is also being used to control upper limb terminal devices, achieving more precise control through sensor acquisition and intent recognition.

[0004] Although human-in-the-loop control strategies have made progress in various fields, some problems still exist in the application of walking-assisting exoskeletons. One of the most important problems is the instability of the instantaneous metabolic rate, which makes it difficult to accurately assess the energy expenditure of the subject. In addition, in order to find the optimal control parameter configuration, it is necessary to collect instantaneous metabolic rates under multiple sets of different parameters, which is a time-consuming and resource-intensive process. Therefore, developing an effective human-in-the-loop optimization strategy that can not only assist the subject in walking but also effectively reduce energy consumption during rehabilitation training is crucial to improving rehabilitation outcomes. Summary of the Invention

[0005] The present invention proposes an in-the-loop adaptive control method and system for a walking-assisting exoskeleton based on energy consumption optimization, which solves the problems of unstable instantaneous metabolic rate and long time-consuming search for optimal control parameters in existing exoskeleton control methods.

[0006] To solve the above technical problems, the present invention provides a method for adaptively controlling a walking-assist exoskeleton human-in-the-loop based on energy consumption optimization, which is characterized by comprising the following steps:

[0007] Step S1: clustering the historical control parameters of the exoskeleton to obtain several groups of typical control parameters, collecting the subject's instantaneous metabolic rate under the several groups of typical control parameters, and fitting the instantaneous metabolic rate to obtain the steady-state metabolic rate;

[0008] Step S2: inputting several groups of typical control parameters and corresponding steady-state metabolic rates into a Bayesian optimization model, wherein the Bayesian optimization model outputs a control parameter combination that minimizes the steady-state metabolic rate as the current optimal control parameter;

[0009] Step S3: obtaining a motion trajectory of the subject under the optimal control parameters, fitting the motion trajectory to obtain an expected trajectory of the subject, driving the exoskeleton according to the expected trajectory, and updating the input of the Bayesian optimization model according to the optimal control parameters and the subject's steady-state metabolic rate at that time;

[0010] Step S4: Repeat steps S2 to S3 until the subject completes the rehabilitation exercise.

[0011] Preferably, in step S1, the instantaneous metabolic rate of the subject is calculated by collecting the oxygen and carbon dioxide contents produced by the subject's breathing, and the expression for calculating the instantaneous metabolic rate is:

[0012] y(t)=0.278*VO2(t)+0.075*VCO2(t);

[0013] Where y(t) is the instantaneous metabolic rate of the subject at time t; VO2(t) and VCO2(t) are the oxygen and carbon dioxide contents collected at time t, respectively.

[0014] Preferably, in step S1, a first-order linear differential equation is used to fit the instantaneous metabolic rate to obtain the steady-state metabolic rate of the subject, and the expression for calculating the steady-state metabolic rate is:

[0015]

[0016] Where y(t)' represents the first derivative of y(t); τ is the time constant; and E is the steady-state metabolic value.

[0017] Preferably, the Bayesian optimization model outputting the current optimal control parameters in step S2 includes the following steps:

[0018] Step S21: using several groups of typical control parameters and corresponding steady-state metabolic rates as the prior distribution of the Bayesian optimization model, and fitting the prior distribution using a Gaussian probability model to obtain the relationship space between the control parameters and the steady-state metabolic rate;

[0019] Step S22: Calculate the expected improvement value of each group of control parameters in the relationship space, and select the control parameter combination with the largest expected improvement value as the optimal control parameter.

[0020] Preferably, the expression of the prior distribution in step S21 is:

[0021] f(x1:n )~N(μ0(x 1:n ),Σ0(x 1:n ,x 1:n ));

[0022]

[0023] Where μ0 represents the mean function; Σ0 represents the kernel function, N() represents the normal distribution; x' represents the center of the kernel function; x is the control parameter of the exoskeleton; and a0 is the parameter.

[0024] Preferably, in step S21, the relationship space is used as the posterior distribution of the Bayesian optimization model, and the expression of the posterior distribution is:

[0025]

[0026] u n (x)=Σ0(x,x 1:n )Σ0(x 1:n ,x 1:n ) -1 (f(x 1:n )-μ0(x 1:n ))+u0(x);

[0027]

[0028] Among them, u n (x) is the posterior mean, is the posterior variance.

[0029] Preferably, the expression for calculating the expected improvement value in step S22 is:

[0030]

[0031] Where θ represents the hyperparameter of the prior distribution; x is the control parameter of the exoskeleton; is the known sample data; E is the steady-state metabolic rate; E best is the optimal observation value of the current iteration; p(|) is the conditional probability; f(x) is the optimization function; n represents the maximum number of sample points in the parameter space; m is one of the sample points.

[0032] Preferably, in step S3, the interpolation method is used to fit the motion trajectory to obtain the subject's expected trajectory, including the following steps: calculating the slope of each data point on the motion trajectory, constructing a piecewise cubic polynomial based on the coordinates of the data point and the corresponding slope, each piecewise cubic polynomial is defined between two adjacent data points, and combining all piecewise cubic polynomials to obtain the expected trajectory. The expression of the expected trajectory is:

[0033]

[0034] Where H3(x) is the desired trajectory; f(x) is the motion trajectory, f'(x) is the derivative of the motion trajectory; x1 and x0 are the coordinates of two adjacent points; x is an arbitrary point in the interpolation interval; a0, a1, β0, and β1 are the parameters of the piecewise cubic polynomial.

[0035] Preferably, in step S3, a PID control algorithm based on gravity compensation is used to drive the exoskeleton.

[0036] The present invention also provides a human-in-the-loop adaptive control system for a walking-assist exoskeleton based on energy consumption optimization, which is applicable to the above-mentioned human-in-the-loop adaptive control method for a walking-assist exoskeleton based on energy consumption optimization, comprising: a data acquisition module, a steady-state metabolic assessment module, a trajectory optimization module, an exoskeleton control module, and an iterative optimization module;

[0037] The data acquisition module collects multi-sensor data from the exoskeleton and the subject and performs data preprocessing. The multi-sensor data includes the joint angles of the exoskeleton and the oxygen and carbon dioxide levels produced during the subject's breathing.

[0038] The steady-state metabolic assessment module calculates the subject's instantaneous metabolic rate using the collected oxygen and carbon dioxide content data, and obtains the steady-state metabolic rate by fitting the instantaneous energy changes during the measurement process using an inverse kinetic model and a first-order linear differential equation;

[0039] The trajectory optimization module includes a parameter clustering module, a Bayesian optimization module and a trajectory generation module;

[0040] The parameter clustering module performs cluster analysis on the historical control parameters of the exoskeleton and identifies several groups of typical control parameters;

[0041] The Bayesian optimization module uses a Bayesian optimization algorithm to find the optimal control parameter combination that minimizes the steady-state metabolic rate based on typical control parameters and corresponding steady-state metabolic rate data;

[0042] The trajectory generation module generates the subject's motion trajectory under the optimal control parameters and fits the trajectory data using an interpolation method to obtain the subject's desired trajectory.

[0043] The exoskeleton control module controls the movement of the exoskeleton robot in real time according to the generated desired trajectory to ensure accurate execution of rehabilitation exercises;

[0044] The iterative optimization module collects the steady-state metabolic rate of the subject under the expected trajectory, feeds the new data back to the Bayesian optimization module, and updates the optimal control parameter combination.

[0045] The benefits of the present invention include at least:

[0046] 1. By clustering and analyzing historical control parameters, representative typical control parameters can be identified, thus avoiding comprehensive testing of all possible parameter combinations and improving the efficiency of data collection;

[0047] 2. Estimating the steady-state metabolic rate by fitting the instantaneous metabolic rate. The obtained steady-state metabolic rate is stable and efficient, and can more accurately reflect the long-term energy expenditure of the subjects under different control parameters;

[0048] 3. The adopted Bayesian optimization model can efficiently search for the optimal solution based on limited data, reducing the large number of experiments that may be required by traditional search methods. In actual use, the control parameters can be dynamically adjusted according to the real-time feedback and metabolic data of the subjects to adapt to the changing needs of the subjects. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention;

[0050] Figure 2 Schematic diagram of estimating steady-state metabolic rate using the inverse kinetics method;

[0051] Figure 3 Schematic diagram of expected trajectories under different control parameters in an embodiment of the present invention;

[0052] Figure 4 Schematic diagram of metabolic rate under control parameter 1 in an embodiment of the present invention;

[0053] Figure 5 Schematic diagram of metabolic rate under control parameter 2 in an embodiment of the present invention;

[0054] Figure 6 Schematic diagram of metabolic rate under control parameter 3 in an embodiment of the present invention;

[0055] Figure 7 Schematic diagram of metabolic rate under control parameter 4 in an embodiment of the present invention;

[0056] Figure 8 Schematic diagram of comparison of metabolic rates under different control parameters in an embodiment of the present invention;

[0057] Figure 9 3D schematic diagram of metabolic rate under different control parameters in an embodiment of the present invention;

[0058] Figure 10 The motion trajectory of the exoskeleton and its parameterization in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts are within the scope of protection of the present invention.

[0060] To address the numerous challenges and uncertainties associated with adjusting exoskeleton parameters, a human-in-the-loop optimization strategy is introduced during the exoskeleton parameter optimization process. This strategy first requires a detailed assessment of the wearer's physiological state, including metrics such as gait characteristics, energy expenditure, and muscle activity. Sensors and algorithms are then used to monitor and analyze these metrics in real time to assess the exoskeleton's impact on the wearer. Based on these assessment results, control parameters are continuously adjusted. For example, parameters controlling the exoskeleton, such as gait pattern, torque output, and joint stiffness, can be dynamically adjusted based on the wearer's individual characteristics and needs to minimize energy consumption and maximize walking assistance. The optimization process searches for the optimal parameter configuration based on a pre-set objective function. Through continuous iteration and evaluation, the optimal parameter combination is ultimately found, enabling the exoskeleton to achieve optimal performance and reduce the subject's energy consumption during walking assistance.

[0061] Based on the human-in-the-loop optimization strategy, the embodiment of the present invention provides a human-in-the-loop adaptive control method for a walking-assist exoskeleton based on energy consumption optimization, such as Figure 1 As shown, the following steps are included:

[0062] Step S1: Cluster the historical control parameters of the exoskeleton to obtain several groups of typical control parameters of the exoskeleton, collect the instantaneous metabolic rates of the subjects under several groups of typical control parameters, and fit these instantaneous metabolic rates to obtain the steady-state metabolic rate.

[0063] Specifically, the historical control parameters of the exoskeleton are collected, and the cluster centers of each type are obtained using a fuzzy clustering algorithm, which are used as the typical control parameters of the exoskeleton.

[0064] Under each set of typical control parameters, multi-sensor data from the exoskeleton and the subject are collected and pre-processed. The multi-sensor data includes the exoskeleton joint angles, the oxygen and carbon dioxide levels VO2 and VCO2 during the subject's breathing. The instantaneous metabolic rate of the subject is calculated based on the collected oxygen and carbon dioxide levels, and the steady-state metabolic rate can be fitted based on the metabolic level in the non-steady state. Figure 2 The figure shows the structure of the inverse dynamics model in the Laplace domain. The inverse dynamics method can be used to convert transient metabolism into steady-state metabolism. The mathematical expression of this process in the Laplace domain is:

[0065] E(s)=H(s)Y(s);

[0066]

[0067] where s represents the Laplace domain; E(s) represents the steady-state metabolic rate; H(s) represents the first-order linear ordinary differential system; Y(s) represents the instantaneous metabolic rate; the parameter τ is the time constant that characterizes the rate of change; A represents the amplitude of the change; and δ is the fixed time delay between the energy consumed by the muscle and the energy measured by the metabolic device.

[0068] In the time domain, the steady-state metabolic rate E under each set of exoskeleton control parameters can be taken as a constant. Under each set of exoskeleton control parameters, a series of n respiratory responses y = [y(1) y(2) … y(n)] are measured, where y(t) = 0.278*VO2(t) + 0.075*VCO2(t), and the corresponding time is t = [t(1) t(2) … t(n)]. The continuous dynamics between the i-th respiratory response y(i+1) of energy expenditure and the steady-state metabolic rate E can be discretized as:

[0069]

[0070] When dt=t(i+1)-t(i), it can be expressed as a system of linear equations:

[0071]

[0072] where y1 is a constant related to the initial respiratory response.

[0073] According to the above formula, we can get the matrix A∈R n×2 When i = 1, the initialization matrix is ​​A(1,1) = 1 and A(1,2) = 0, then when i∈[2,3,…,n], the iteration matrix can be constructed as:

[0074]

[0075] A is an n×2 matrix, where the elements are arranged as follows:

[0076]

[0077] The steady-state metabolic rate can be calculated as follows:

[0078]

[0079] Among them, A + is the pseudo-inverse of A, and is an estimate of the initial respiratory response and steady-state metabolic value E.

[0080] Finally, the temporal relationship between y and E can be established:

[0081]

[0082] Where E is the steady-state metabolic rate; y(t) is the instantaneous metabolic rate of the subject at time t, y(t)' is the first-order derivative of y(t); and τ is the time constant that characterizes the rate of change.

[0083] Step S2: Several groups of typical control parameters and corresponding steady-state metabolic rates are input into the Bayesian optimization model, and the Bayesian optimization model outputs the control parameter combination that minimizes the steady-state metabolic rate as the current optimal control parameter.

[0084] like Figure 3 The following are the expected motion trajectories of the subjects under four typical control parameter sets. These control parameters are mainly responsible for adjusting the maximum angle of the trajectory and the time of the power assist rise. The embodiment of the present invention ensures that each gait cycle lasts the same time in all cases. Figure 3 It can be seen that the trajectory amplitude is the largest and the boost rise time is the longest under control parameter 3, the boost rise time is the shortest under control parameter 2, and the trajectory amplitude is the smallest under control parameter 4. The expected trajectories under the four typical control parameter groups obtained in the embodiment of the present invention are shown in Table 1.

[0085] Table 1 Expected trajectories under different control parameters

[0086] Control parameter 1 Control parameter 2 Control parameter 3 Control parameter 4 Maximum angle(°) 40 40 50 30 Rise time (%) 0.35 0.15 0.40 0.20 Fall time (%) 0.40 0.20 0.45 0.25

[0087] like Figures 4 to 7 The following are the metabolic rate measurements of the subjects while walking under four different control parameters. The gray line in the figure is the curve obtained after steady-state fitting. From the steady-state metabolic rate, it can be seen that the metabolic rate under control parameter 2 is significantly lower than that under the other three control parameters, and there is no significant difference in the metabolic rate under control parameters 1, 3, and 4. Figure 8 The steady-state metabolic rate under different conditions was calculated by averaging the steady-state data. Under control parameter 2, the exoskeleton's maximum angle was 40°, the assist time was 0.15, and its lowest metabolic rate was 89.9 / 60. Under control parameter 3, the exoskeleton's maximum angle was 50°, the assist time was 0.4, and its highest metabolic rate was 110.7 / 60.

[0088] By analyzing the metabolic rate under these known control parameters, a model such as Figure 9 The three-dimensional model shown in the figure is combined with the Bayesian optimization algorithm to show the general trend of metabolic rate changes with different control parameters, and then the optimal control parameters can be obtained for the subjects. Figure 9 The lowest point of the three-dimensional surface is taken as the optimal control parameter of the subject.

[0089] In the embodiments of the present invention, a Bayesian optimization model is used to implement this process. Specifically, several sets of typical control parameters and their corresponding steady-state metabolic rates are used as the prior distributions of the Bayesian optimization model. The proxy model of the Bayesian optimization model adopts a Gaussian probability model, and the acquisition function adopts an expected improvement function. The Bayesian optimization model calculates its posterior distribution based on the prior distribution and then uses the expected improvement function to calculate the optimal control parameters.

[0090] Assuming that the steady-state metabolic rates corresponding to n sets of control parameters have been obtained, the prior distribution of the Gaussian probability model is as follows:

[0091] f(x 1:n )~N(μ0(x 1:n ),Σ0(x 1:n ,x 1:n ));

[0092] Among them, μ0 represents the average function; Σ0 represents the kernel function; N(μ0(x 1:n ),Σ0(x 1:n ,x 1:n )) represents a normal distribution.

[0093] The expression of kernel function Σ0 is as follows:

[0094]

[0095] Where x' represents the center of the kernel function; x represents a typical control parameter; and a0 represents a parameter.

[0096] The posterior distribution obtained based on the first n results is as follows:

[0097]

[0098] u n (x)=Σ0(x,x 1:n )Σ0(x 1:n ,x 1:n ) -1 (f(x 1:n )-μ0(x 1:n ))+u0(x);

[0099]

[0100] Among them, u n (x) is the posterior mean; is the posterior variance.

[0101] The expression of the expected improvement function used is:

[0102]

[0103] Where θ represents the hyperparameter of the prior distribution; x is the control parameter of the exoskeleton; is the known sample data; E is the steady-state metabolic rate; E best is the optimal observation value of the current iteration; p(|) is the conditional probability; f(x) is the optimization function; n represents the maximum number of sample points in the parameter space; m is one of the sample points.

[0104] Step S3: Obtain the subject's motion trajectory under the optimal control parameters, fit the motion trajectory to obtain the subject's expected trajectory, drive the exoskeleton according to the expected trajectory, and update the input of the Bayesian optimization model based on the optimal control parameters and the subject's steady-state metabolic rate at this time.

[0105] like Figure 10 As shown, in the embodiment of the present invention, the PCHIP interpolation method is used to fit the subject's motion trajectory to obtain the expected trajectory. By calculating the slope of each data point on the motion trajectory, a piecewise cubic polynomial is constructed based on the coordinates of the data point and the corresponding slope. Each piecewise cubic polynomial is defined between two adjacent data points. All piecewise cubic polynomials are combined to obtain the expected trajectory. The use of the PCHIP interpolation method ensures that the curve generated within adjacent data points is monotonic, which is more consistent with the actual human motion trajectory. The expression of the expected trajectory is:

[0106]

[0107] Where H3(x) is the desired trajectory; f(x) is the motion trajectory, f'(x) is the derivative of the motion trajectory; x1 and x0 are the coordinates of two adjacent points; x is an arbitrary point in the interpolation interval; a0, a1, β0, and β1 are the parameters of the piecewise cubic polynomial.

[0108] After inputting the desired trajectory into the exoskeleton's position controller, sensors are used to collect the subject's metabolic rate and corresponding control parameters during movement, which serve as input for the next iteration of the Bayesian optimization algorithm.

[0109] Specifically, in an embodiment of the present invention, a PID control algorithm based on gravity compensation is used to accurately control the exoskeleton robot, thereby driving the subject to perform rehabilitation exercises. If the subject is standing normally and the scale lines of the three modules of the hip, knee and ankle are aligned with zero, it can be assumed to be a zero torque state. If the exoskeleton starts to take a step, a torque that balances gravity must be generated in order to maintain balance during the walking phase. After compensating for gravity, the exoskeleton will not fall due to the influence of gravity.

[0110] Step S4: Repeat steps S2 to S3 until the subject completes the rehabilitation exercise.

[0111] The embodiment of the present invention proposes a method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization. It uses multi-sensor technology to collect multimodal data, which not only enriches the data type but also improves the accuracy of the data, providing a solid foundation for subsequent energy consumption analysis and control. The instantaneous metabolic cost is estimated by the inverse model method, and the relationship between the instantaneous energy cost and the measurement cost is constructed as a first-order linear differential equation, which effectively fits the instantaneous energy change during the measurement process, thereby achieving stable and efficient acquisition of steady-state metabolic values. The Bayesian algorithm is used to optimize the exoskeleton, using a proxy function instead of the objective function, and using the posterior distribution obtained from the previous iteration as prior knowledge in each iteration to optimize the next set of parameters. This method greatly reduces the time cost required to find the optimal solution. The PCHIP interpolation method is used to generate a smooth trajectory that conforms to the characteristics of human motion. This method fits the expected trajectory of the knee joint motion based on known discrete data points, ensuring a smooth transition of the curve between adjacent data points, which is closer to real human motion. A PID control algorithm based on gravity compensation is used to compensate for gravity, ensuring that the exoskeleton will not be adversely affected by gravity during movement. This allows the exoskeleton robot to be precisely controlled and drive the subjects to perform effective rehabilitation exercises.

[0112] An embodiment of the present invention also provides a human-in-the-loop adaptive control system for a walking-assisting exoskeleton based on energy consumption optimization, comprising a data acquisition module, a steady-state metabolic assessment module, a trajectory optimization module, an exoskeleton control module, and an iterative optimization module.

[0113] The data acquisition module is used to collect multi-sensor data from the exoskeleton and the subject and perform data preprocessing. The multi-sensor data includes the joint angles of the exoskeleton and the oxygen and carbon dioxide content produced during the subject's breathing.

[0114] The steady-state metabolic assessment module uses the collected oxygen and carbon dioxide content data to calculate the subject's instantaneous metabolic rate, and obtains the steady-state metabolic rate by fitting the instantaneous energy changes during the measurement process through an inverse kinetic model and a first-order linear differential equation.

[0115] The trajectory optimization module includes a parameter clustering module, a Bayesian optimization module, and a trajectory generation module. The parameter clustering module performs cluster analysis on the exoskeleton's historical control parameters and identifies several groups of typical control parameters. The Bayesian optimization module uses a Bayesian optimization algorithm to find the optimal control parameter combination that minimizes the steady-state metabolic rate based on the typical control parameters and the corresponding steady-state metabolic rate data. The trajectory generation module generates the subject's motion trajectory under the optimal control parameters and uses interpolation to fit the trajectory data to obtain the subject's desired trajectory.

[0116] The exoskeleton control module controls the movement of the exoskeleton robot in real time according to the generated desired trajectory, ensuring accurate execution of rehabilitation exercises.

[0117] The iterative optimization module is used to collect the steady-state metabolic rate of the subjects under the desired trajectory, feed the new data back to the Bayesian optimization module, and update the optimal control parameter combination.

[0118] The technical features of the above embodiments may be combined in any manner. To simplify the description, not all possible combinations of the technical features in the above embodiments are described. Only preferred embodiments of the present invention are presented. While the description is relatively specific and detailed, it should not be construed as limiting the scope of the present invention. As long as there are no contradictions in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] It should be noted that those skilled in the art may make various modifications and improvements without departing from the scope of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be based on the appended claims.

Claims

1. A human-in-the-loop adaptive control method for a walking-assist exoskeleton based on energy consumption optimization, characterized in that: The following steps are involved: Step S1: clustering the historical control parameters of the exoskeleton to obtain several groups of typical control parameters, collecting the subject's instantaneous metabolic rate under the several groups of typical control parameters, and fitting the instantaneous metabolic rate to obtain the steady-state metabolic rate; Step S2: inputting several groups of typical control parameters and corresponding steady-state metabolic rates into a Bayesian optimization model, wherein the Bayesian optimization model outputs a control parameter combination that minimizes the steady-state metabolic rate as the current optimal control parameter; Step S3: obtaining a motion trajectory of the subject under the optimal control parameters, fitting the motion trajectory to obtain an expected trajectory of the subject, driving the exoskeleton according to the expected trajectory, and updating the input of the Bayesian optimization model according to the optimal control parameters and the subject's steady-state metabolic rate at that time; Step S4: Repeat steps S2 to S3 until the subject completes the rehabilitation exercise.

2. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 1, characterized in that: In step S1, the instantaneous metabolic rate of the subject is calculated by collecting the oxygen and carbon dioxide content produced by the subject's breathing. The expression for calculating the instantaneous metabolic rate is: y(t)=0.278*VO2(t)+0.075*VCO2(t); Where y(t) is the instantaneous metabolic rate of the subject at time t; VO2(t) and VCO2(t) are the oxygen and carbon dioxide contents collected at time t, respectively.

3. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 2, characterized in that: In step S1, a first-order linear differential equation is used to fit the instantaneous metabolic rate to obtain the subject's steady-state metabolic rate. The expression for calculating the steady-state metabolic rate is: Where y(t)' represents the first derivative of y(t); τ is the time constant; and E is the steady-state metabolic value.

4. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 1, characterized in that: The Bayesian optimization model outputting the current optimal control parameters in step S2 includes the following steps: Step S21: using several groups of typical control parameters and corresponding steady-state metabolic rates as the prior distribution of the Bayesian optimization model, and fitting the prior distribution using a Gaussian probability model to obtain the relationship space between the control parameters and the steady-state metabolic rate; Step S22: Calculate the expected improvement value of each group of control parameters in the relationship space, and select the control parameter combination with the largest expected improvement value as the optimal control parameter.

5. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 4, characterized in that: The expression of the prior distribution in step S21 is: f(x 1:n )~N(μ0(x 1:n ),Σ0(x 1:n ,x 1:n )); Where n is the total number of groups of typical control parameters; μ0 represents the average function; Σ0 represents the kernel function, N() represents the normal distribution; x' represents the center of the kernel function; x is the control parameter of the exoskeleton; and a0 is the parameter.

6. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 5, characterized in that: In step S21, the relationship space is used as the posterior distribution of the Bayesian optimization model. The expression of the posterior distribution is: u n (x)=Σ0(x,x 1:n )Σ0(x 1:n ,x 1:n ) -1 (f(x 1:n )-μ0(x 1:n ))+u0(x); Among them, u n (x) is the posterior mean, is the posterior variance.

7. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 4, characterized in that: The expression for calculating the expected improvement value in step S22 is: Where θ represents the hyperparameter of the prior distribution; x is the control parameter of the exoskeleton; is the known sample data; E is the steady-state metabolic rate; E best is the optimal observation value of the current iteration; p(|) is the conditional probability; f(x) is the optimization function; n represents the maximum number of sample points in the parameter space; m is one of the sample points.

8. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 1, characterized in that: In step S3, the motion trajectory is fitted using an interpolation method to obtain the subject's expected trajectory, including the following steps: calculating the slope of each data point on the motion trajectory, constructing a piecewise cubic polynomial based on the coordinates of the data points and the corresponding slope, each piecewise cubic polynomial is defined between two adjacent data points, and combining all piecewise cubic polynomials to obtain the expected trajectory. The expression of the expected trajectory is: Where H3(x) is the desired trajectory; f(x) is the motion trajectory, f'(x) is the derivative of the motion trajectory; x1 and x0 are the coordinates of two adjacent points; x is an arbitrary point in the interpolation interval; a0, a1, β0, and β1 are the parameters of the piecewise cubic polynomial.

9. The method for adaptive control of a walking-assist exoskeleton based on energy consumption optimization according to claim 1, characterized in that: In step S3, a PID control algorithm based on gravity compensation is used to drive the exoskeleton.

10. A human-in-the-loop adaptive control system for a walking-assist exoskeleton based on energy consumption optimization, applicable to the human-in-the-loop adaptive control method for a walking-assist exoskeleton based on energy consumption optimization according to any one of claims 1 to 9, characterized in that: It includes data acquisition module, steady-state metabolic assessment module, trajectory optimization module, exoskeleton control module and iterative optimization module; The data acquisition module collects multi-sensor data from the exoskeleton and the subject and performs data preprocessing. The multi-sensor data includes the joint angles of the exoskeleton and the oxygen and carbon dioxide levels produced during the subject's breathing. The steady-state metabolic assessment module calculates the subject's instantaneous metabolic rate using the collected oxygen and carbon dioxide content data, and obtains the steady-state metabolic rate by fitting the instantaneous energy changes during the measurement process using an inverse kinetic model and a first-order linear differential equation; The trajectory optimization module includes a parameter clustering module, a Bayesian optimization module and a trajectory generation module; The parameter clustering module performs cluster analysis on the historical control parameters of the exoskeleton and identifies several groups of typical control parameters; The Bayesian optimization module uses a Bayesian optimization algorithm to find the optimal control parameter combination that minimizes the steady-state metabolic rate based on typical control parameters and corresponding steady-state metabolic rate data; The trajectory generation module generates the subject's motion trajectory under the optimal control parameters and fits the trajectory data using an interpolation method to obtain the subject's desired trajectory. The exoskeleton control module controls the movement of the exoskeleton robot in real time according to the generated desired trajectory to ensure accurate execution of rehabilitation exercises; The iterative optimization module collects the steady-state metabolic rate of the subject under the expected trajectory, feeds the new data back to the Bayesian optimization module, and updates the optimal control parameter combination.

Images