Gait optimization method for foot drop exoskeleton based on multi-muscle coordination

Abstract

The present application relates to the field of intelligent medical technology, especially to a foot drop exoskeleton gait optimization method based on multi-muscle coordination, which comprises collecting multi-channel surface electromyography signals related to ankle dorsiflexion and plantarflexion of a user through a surface electromyography sensor, pre-processing the multi-channel surface electromyography signals to obtain the root mean square (RMS) of the electromyography of each target muscle in the gait cycle; based on the root mean square (RMS) of the electromyography, calculating the mutual information (MI) between each target muscle and the exoskeleton assistance parameter, and determining the weight of each target muscle according to the mutual information (MI) to construct a multi-muscle weighted cost function; feeding back the calculation result of the multi-muscle weighted cost function to the control end of the foot drop exoskeleton, iteratively optimizing the exoskeleton assistance parameter through a Bayesian optimization algorithm until convergence, obtaining the optimal exoskeleton assistance parameter adapted to the user, and completing the gait optimization of the foot drop exoskeleton. The naturalness of the human-machine interaction of the exoskeleton is significantly improved, and the gait compensation risk is reduced.

Description

Technical Field

[0001] This application relates to the field of smart medical technology, and in particular to a gait optimization method for foot drop exoskeleton based on multi-muscle coordination. Background Technology

[0002] Stroke, a leading neurological disorder causing chronic disability and death worldwide, is fundamentally caused by localized nerve damage resulting from cerebral artery occlusion or rupture. Patients often experience sensory, cognitive, and motor dysfunction, with foot drop being the most common type of gait disorder. Essentially, foot drop is the loss or limitation of ankle dorsiflexion, preventing patients from lifting their forefoot during the load response phase (initial support) and swing phase of the gait cycle. This manifests as typical symptoms such as foot slapping and dragging, severely reducing daily living abilities and significantly increasing the risk of falls, placing a heavy burden on patients and their families.

[0003] To improve gait function in patients with foot drop, clinicians and industry have developed two core types of rehabilitation assistive devices: passive ankle-foot orthoses and active wearable exoskeletons for foot drop. Passive ankle-foot orthoses restrict the range of motion of the ankle joint through mechanical structures (such as elastic supports and rigid splints), directly preventing foot drop and reducing foot-slapping. They have the advantages of simple structure and low cost. However, these devices can only provide passive support and cannot adapt to the patient's active movement intentions. Long-term use can easily lead to insufficient activity of dorsiflexion-related muscles (such as the tibialis anterior muscle) and even inhibit the spontaneous recovery of neuromuscular function, making it difficult to meet the core needs of active compensation and functional reconstruction in rehabilitation training.

[0004] With the integration of robotics and biomedical engineering, active wearable foot drop exoskeletons have become a research hotspot and mainstream development direction in recent years. These devices, powered by active sources such as motors, pneumatic / hydraulic actuators, can provide controllable dorsiflexion assistance to the ankle joint during key gait phases. Clinical studies have shown that they can effectively increase foot clearance, improve lower limb kinematic parameters (such as ankle joint range of motion and stride symmetry), and reduce energy expenditure during walking, significantly outperforming the rehabilitation effects of passive orthotics.

[0005] However, existing active foot drop exoskeletons mostly rely on predefined trajectory designs. This involves collecting gait data from healthy individuals, pre-setting a fixed ankle dorsiflexion assistance curve (such as fixed timing and magnitude of assistance), and applying this trajectory to all users. However, foot drop patients exhibit significant individual differences in the degree of neuromuscular damage (such as muscle strength grading and spasticity) and residual motor function. For example, patients with mild damage may require only a small amount of assistance to complete dorsiflexion, while those with severe damage require greater assistance and earlier timing. Using a uniform predefined trajectory can lead to a mismatch between the assistance force and the patient's active movement intention (e.g., assistance too early or too late), or even human-machine interaction conflicts (e.g., exoskeleton assistance antagonizing the patient's residual muscle strength), thereby disrupting the natural gait pattern and increasing the risk of gait compensation (e.g., relying on hip elevation to compensate for foot drop).

[0006] Currently, personalized gait generation technologies for ankle dorsiflexion rely on a cost function as the core feedback mechanism. This cost function quantifies the effects of human-computer interaction (such as muscle load and energy consumption) to guide the exoskeleton in adjusting auxiliary parameters. In related technologies, the cost function design typically incorporates only two types of physiological indicators:

[0007] Metabolic energy consumption: Estimated by collecting respiratory data for more than 2 minutes using respiratory gas analysis equipment (such as a metabolic cart). Although it can reflect the overall exercise energy consumption, the equipment is large and bulky, which can easily cause discomfort to patients, and it cannot provide real-time feedback.

[0008] Single muscle activation: The activity level of a dorsiflexor muscle (such as the tibialis anterior) is calculated based on surface electromyography (sEMG) signals. This ignores the multi-muscle synergy of ankle joint movement, resulting in a one-sided feedback signal that cannot accurately reflect the true synergistic state of human-computer interaction.

[0009] The aforementioned defects directly lead to insufficient accuracy of the cost function in responding to the human-computer interaction state, resulting in poor individual adaptability and low optimization efficiency of the adaptive optimization strategy, making it difficult to find auxiliary parameters that are truly suitable for the patient. Summary of the Invention

[0010] To address the aforementioned technical issues, this application provides a gait optimization method for foot drop exoskeleton based on multi-muscle coordination.

[0011] Firstly, this application provides a gait optimization method for foot drop exoskeleton based on multi-muscle coordination:

[0012] S1. After the foot drop exoskeleton applies initial ankle dorsiflexion assistance to the user, it collects multi-channel surface electromyography (EMG) signals related to ankle dorsiflexion and plantar flexion through surface EMG sensors. The multi-channel EMG signals include at least the EMG signals of the tibialis anterior, soleus, and gastrocnemius muscles.

[0013] S2. Preprocess the multi-channel surface electromyography signals to obtain the root mean square (RMS) electromyography of each target muscle during the gait cycle.

[0014] S3. Based on the root mean square electromyography (RMS), calculate the mutual information (MI) between each target muscle and the exoskeleton auxiliary parameters, and determine the weight of each target muscle according to the mutual information MI, and construct a multi-muscle weighted cost function.

[0015] S4. Feedback the calculation results of the multi-muscle weighted cost function to the control end of the foot drop exoskeleton, and iteratively optimize the exoskeleton auxiliary parameters through the Bayesian optimization algorithm until convergence, thereby obtaining the optimal exoskeleton auxiliary parameters adapted to the user and completing the gait optimization of the foot drop exoskeleton.

[0016] By adopting the above technical solution, and collecting multi-channel electromyographic signals from the tibialis anterior, soleus, and gastrocnemius muscles, the limitations of current methods that rely solely on a single muscle or predefined trajectory are overcome. Furthermore, by combining mutual information to construct a multi-muscle weighted cost function, a comprehensive quantification of human-machine collaboration is achieved, avoiding the one-sidedness of feedback from a single indicator. Finally, by using Bayesian optimization to rapidly iterate and optimize auxiliary parameters, the problems of slow convergence and poor individual adaptability of traditional adaptive methods are solved. Overall, a closed loop of multi-muscle collaboration assessment and personalized parameter optimization is realized, which can generate gait assistance strategies adapted to the neuromuscular functions of different patients, significantly improving the naturalness of exoskeleton-human-machine interaction and rehabilitation effects.

[0017] Optionally, the preprocessing in step S2 includes the following sub-steps:

[0018] S21. The heel contact event is detected by the inertial measurement unit (IMU) at the toe of the foot drop exoskeleton, and the multi-channel surface electromyography signal is divided according to the gait cycle based on the heel contact event.

[0019] S22. The segmented electromyographic signals are sequentially subjected to high-pass filtering, full-wave rectification, and low-pass filtering. The high-pass filtering is a second-order Butterworth filter with a cutoff frequency of 20Hz, and the low-pass filtering is a second-order Butterworth filter with a cutoff frequency of 10Hz.

[0020] S23. Calculate the root mean square (RMS) of electromyography (EMG) signals after filtering. In the process of calculating the RMS, the original EMG data in each gait cycle is uniformly resampled to several sampling points, and the RMS is calculated based on the amplitude of the several sampling points.

[0021] By adopting the above technical solution, the gait cycle of electromyography (EMG) signals is accurately segmented by detecting heel-to-ground events using an IMU, ensuring that subsequent analysis matches the actual gait stages. Simultaneously, a second-order Butterworth filter combination of a 20Hz high-pass filter and a 10Hz low-pass filter effectively filters out motion artifacts (such as limb swaying noise) and high-frequency interference in the EMG signals, preserving the true envelope information of muscle activation. Furthermore, the EMG data for each gait cycle are uniformly resampled to multiple points, avoiding inconsistencies in the number of sampling points caused by differences in stride length across different gaits. This ensures the standardization and accuracy of the root mean square (RMS) calculation of EMG signals, providing high-quality data support for subsequent cost function construction and resolving the low reliability of existing EMG signal preprocessing methods.

[0022] Optionally, step S2 further includes a sub-step S20 for determining the stability of the multi-channel surface electromyography signal, specifically including:

[0023] S201. Calculate the average RMS of the continuously acquired electromyographic signals from step 1 to step i, denoted as A. i (i≥1), where A i This is the arithmetic mean of the electromyographic RMS from step 1 to step i;

[0024] S202. Calculate the rate of change ΔA of adjacent average RMS using the forward difference method. i =A i+1 -A i Combine all ΔAi from step 1 to step i into a rate of change vector;

[0025] S203. Divide the rate of change vector into multiple sliding windows B using a sampling rate of 1Hz and a window size of s=5. n , where n is the window number;

[0026] S204, For adjacent sliding windows B n With B n-1 Perform a two-sample z-test, with the hypothesis H0: μ(B) n )=μ(B n-1 μ is the mean of the data within the window; when the p-value of the two-tailed test is ≥0.95, it is determined that the multi-channel surface electromyography signal has reached stability, the acquisition is stopped and step S21 is executed.

[0027] By adopting the above technical solution, the method calculates the average RMS, extracts the rate of change vector, and performs a sliding window z-test to accurately identify the steady state of electromyographic signals with a confidence level of p≥0.95, without waiting for a lengthy fixed acquisition time. This method can shorten the acquisition time of stable signals to about 1 minute (55 steps), which significantly reduces the time cost compared with existing technologies. It not only reduces the burden of continuous walking for patients, but also achieves rapid and effective acquisition of electromyographic signals, laying the foundation for rapid iteration of subsequent weighted cost function calculation and Bayesian optimization, and solving the bottleneck of lagging human-computer interaction feedback.

[0028] Optionally, the process of calculating mutual information MI in step S3 includes:

[0029] The root mean square (RMS) of stable electromyography collected under various preset exoskeleton-assisted modes is arithmetically averaged. Mutual information (MI) is calculated based on the joint probability density p(x,y) and marginal probability densities p(x) and p(y) of the averaged RMS and the corresponding exoskeleton-assisted parameters; where x is the averaged RMS and y is the exoskeleton-assisted parameter.

[0030] The determination of the weights of each target muscle includes: calculating the mutual information (MI) of each target muscle. i Normalized to the interval 0 to 1, weight W i =MI i / ΣMI i , where ΣMI i It is the sum of the mutual information of the tibialis anterior, soleus and gastrocnemius muscles, and the weights w1 of the tibialis anterior, w2 of the soleus and w3 of the gastrocnemius are all positive numbers.

[0031] The multi-muscle weighted cost function is CF=w1×TA + w2×SOL +w3×GAS, where TA is the average RMS of the tibialis anterior muscle, SOL is the average RMS of the soleus muscle, and GAS is the average RMS of the gastrocnemius muscle. TA, SOL, and GAS are all normalized using the average maximum muscle activation value under unassisted mode.

[0032] By adopting the above technical solution, mutual information is calculated by pre-setting the RMS mean values ​​of multiple assistance modes, amplifying the differences in muscle activation between different assistance modes, and improving the accuracy of mutual information in representing the dependence of muscles on assistance parameters. At the same time, the weights of the tibialis anterior, soleus, and gastrocnemius muscles are all constrained to positive numbers, avoiding the problem of abnormally high activation of antagonist muscles caused by traditional negative weight design, balancing the activation levels of prime movers and antagonists, which is more in line with the biomechanical coordination law of the ankle joint and significantly reduces the risk of gait compensation. The final constructed cost function can comprehensively reflect the correlation between the multi-muscle synergy state and the exoskeleton assistance effect, solving the problem of one-sided feedback and easy gait deformation of existing cost functions, and improving the feedback accuracy of adaptive optimization.

[0033] Optionally, the Bayesian optimization algorithm described in step S4 includes the following sub-steps:

[0034] S41. Initialize the posterior distribution: Randomly select multiple predefined exoskeleton auxiliary parameters from uniform intervals, evaluate the multi-muscle weighted cost function under each parameter, and construct the initial posterior distribution of the objective function based on the evaluation results of each group.

[0035] S42. Constructing a surrogate model: A Gaussian process (GP) is used as a surrogate model to model the distribution of the objective function. The covariance function of the Gaussian process is a combination of a Matern kernel and a White kernel with a smoothness parameter of 2.5. The Matern kernel is used to capture the smooth underlying structure of the objective function, and the White kernel is used to model the observation noise, which is the unstructured Gaussian noise in the electromyography signal acquisition process.

[0036] S43. Determine the next sampling point: Using the expected improved EI as the acquisition function, calculate the EI value of the candidate auxiliary parameters. The EI value is calculated based on the predicted mean μ(x) and predicted standard deviation σ(x) of the Gaussian process output. The next evaluation point is determined by maximizing the EI value. Furthermore, multiple random restarts are performed throughout the search space to avoid getting trapped in local optima;

[0037] S44, Iterative Optimization: Move to the next evaluation point The corresponding exoskeleton auxiliary parameters are applied to the foot drop exoskeleton. Steps S1-S3 are repeated to obtain a new multi-muscle weighted cost function result. The surrogate model is updated based on the new result, and step S43 is executed again to determine the new... The iteration continues until a preset stopping condition is met, at which point the iteration stops and the current iteration is output. As the optimal auxiliary parameters for the exoskeleton.

[0038] By adopting the above technical solutions, the posterior distribution is initialized with multiple predefined parameters, reducing the dependence on the number of initial samples and lowering the optimization threshold. A Gaussian process surrogate model combining Matern kernel and White kernel is used. The Matern kernel can accurately capture the smooth relationship between auxiliary parameters and muscle activation, while the White kernel effectively models unstructured noise in electromyography signals, improving the robustness of the surrogate model to complex human-computer interaction scenarios. By combining the expected improvement (EI) acquisition function and the multiple random restart strategy, it can both prioritize the exploration of potential optimal regions (utilization) and avoid missing unexplored regions (exploration), while preventing getting trapped in local optima, significantly reducing the number of optimization iterations, and achieving rapid global optimization of exoskeleton auxiliary parameters. This solves the problems of low efficiency and poor reliability of optimization results in traditional adaptive methods.

[0039] Optionally, the exoskeleton assist parameters are two-dimensional parameters, including peak force and peak time; the peak force is the maximum value of the ankle dorsiflexion assist force provided by the exoskeleton; and the peak time is the moment when the exoskeleton provides the maximum assist force.

[0040] By adopting the above technical solution, the exoskeleton assist parameters are defined as two-dimensional parameters of peak force (N / kg) + peak time (% gait cycle), which accurately quantifies the magnitude and timing of the assist force and meets the different dorsiflexion requirements of the ankle joint during the load response period and swing period.

[0041] Optionally, step S4 further includes: converting the optimal exoskeleton assistance parameters into an assist trajectory in units of gait cycles using cubic spline interpolation technology, wherein the assist trajectory is transmitted to the user's ankle joint through the cordless drive system of the foot drop exoskeleton to achieve personalized gait assistance.

[0042] By adopting the above technical solution, the optimized parameters are converted into an assist trajectory in gait cycle units through cubic spline interpolation, ensuring that the assist changes are continuous and smooth, conforming to the force change law of the human body's natural gait. Finally, the trajectory is transmitted through a cordless drive system, avoiding the restriction of gait by wired drive, further improving the naturalness and comfort of the assistance, solving the problems of rigidity of existing predefined trajectories and mismatch with individual gait cycles, and realizing the accurate implementation of personalized assist trajectories.

[0043] Optionally, the surface electromyography (EMG) sensor mentioned in step S1 is a wireless surface EMG sensor with a sampling rate of 2000Hz; the EMG signal collected by the surface EMG sensor is transmitted to the computer via a serial port, and the computer aligns the EMG signal with the foot drop exoskeleton master control terminal through a socket synchronization flag to ensure that the EMG signal is synchronized with the time of the exoskeleton assisted movement.

[0044] By adopting the above technical solution, the timing of electromyographic signals and exoskeleton motion data is aligned through serial port transmission and socket synchronization flags, avoiding mismatch between auxiliary parameters and muscle state caused by signal delay, and ensuring the accuracy of the timing correlation of subsequent multi-muscle collaborative analysis and cost function calculation.

[0045] Secondly, this application provides a foot drop exoskeleton gait optimization device based on multi-muscle coordination; the device includes modules for performing the foot drop exoskeleton gait optimization method based on multi-muscle coordination in the first aspect or any possible implementation of the first aspect:

[0046] This foot drop exoskeleton gait optimization device based on multi-muscle coordination includes:

[0047] The acquisition module is used to acquire multi-channel surface electromyography (EMG) signals related to ankle dorsiflexion and plantar flexion after the foot drop exoskeleton applies initial ankle dorsiflexion assistance to the user. The multi-channel EMG signals include at least the EMG signals of the tibialis anterior, soleus, and gastrocnemius muscles.

[0048] The preprocessing module is used to preprocess the multi-channel surface electromyography signals to obtain the root mean square (RMS) electromyography of each target muscle during the gait cycle.

[0049] The cost construction module is used to calculate the mutual information (MI) between each target muscle and the exoskeleton auxiliary parameters based on the root mean square (RMS) electromyography, and to determine the weight of each target muscle according to the mutual information MI, and to construct a multi-muscle weighted cost function.

[0050] The optimization module is used to feed back the calculation results of the multi-muscle weighted cost function to the control end of the foot drop exoskeleton, and iteratively optimize the auxiliary parameters of the exoskeleton through a Bayesian optimization algorithm until convergence, thereby obtaining the optimal auxiliary parameters of the exoskeleton that are suitable for the user and completing the gait optimization of the foot drop exoskeleton.

[0051] Thirdly, this application provides a computer device for a foot drop exoskeleton, including a processor, a memory, and a communication bus. The communication bus is used to establish a communication connection between the processor and the memory. The processor is used to execute a computer program stored in the memory to implement the foot drop exoskeleton gait optimization method based on multi-muscle coordination as described in any of the preceding claims.

[0052] Fourthly, this application also provides a computer-readable storage medium storing a computer program; the computer program can be executed by a processor to implement the foot drop exoskeleton gait optimization method based on multi-muscle coordination as described above.

[0053] Fifthly, this application also provides a computer program product, including a computer program that can be executed by a processor to implement the foot drop exoskeleton gait optimization method based on multi-muscle coordination as described above.

[0054] This application includes at least the following beneficial technical effects:

[0055] Addressing the core pain points of existing foot drop exoskeletons—namely, the lack of personalized assistance, insufficient multi-muscle synergy assessment, and low optimization efficiency—this study collects multi-channel electromyographic signals from key ankle dorsiflexion and plantarflexion muscle groups (tibialis anterior, soleus, and gastrocnemius). By combining this with mutual information to construct a multi-muscle positive weighted cost function, it overcomes the limitations of single-muscle indices or negative weighted designs for antagonistic muscles, accurately quantifying the true state of human-machine collaboration. Simultaneously, relying on Bayesian optimization, it achieves efficient iterative convergence of auxiliary parameters. This not only generates gait assistance strategies tailored to the individual functions of foot drop patients with varying degrees of neuromuscular injury, significantly improving the naturalness of exoskeleton-human interaction and reducing gait compensation risks, but also greatly enhances the efficiency of personalized gait optimization, providing precise and efficient technical support for foot drop rehabilitation. Attached Figure Description

[0056] Figure 1 A schematic diagram of a gait optimization method for foot drop based on multi-muscle coordination using an exoskeleton, provided in an embodiment of this application;

[0057] Figure 2 A schematic diagram of the physiological anatomy of the lower limb muscles provided in this application embodiment;

[0058] Figure 3 A schematic diagram of a Bayesian optimization framework based on a weighted cost function is provided for embodiments of this application;

[0059] Figure 4 This application provides a schematic diagram of exoskeleton-assisted force modeling.

[0060] Figure 5 A schematic diagram of a gait optimization device for foot drop exoskeleton based on multi-muscle coordination provided in an embodiment of this application;

[0061] Figure 6 This is a schematic diagram of a computer device structure for a foot drop exoskeleton, provided as an embodiment of this application. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0063] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items. The term “exemplary” means “serving as an example, embodiment, or illustration,” and any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. The terms “first” and “second” are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as “first” or “second” may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, “a plurality” means two or more.

[0064] To address the shortcomings and improvement needs in the aforementioned personalized gait generation and cost function design techniques, this application provides a gait optimization method based on a multi-muscle coordination exoskeleton for foot drop; this method can solve the aforementioned technical problems.

[0065] refer to Figure 1 A gait optimization method based on a multi-muscle synergy exoskeleton for foot drop mainly includes the following steps:

[0066] S1. After the foot drop exoskeleton applies initial ankle dorsiflexion assistance to the user, it collects multi-channel surface electromyography (EMG) signals related to ankle dorsiflexion and plantar flexion through surface EMG sensors. The multi-channel EMG signals include at least the EMG signals of the tibialis anterior, soleus, and gastrocnemius muscles.

[0067] S2. Preprocess the multi-channel surface electromyography signals to obtain the root mean square (RMS) electromyography of each target muscle during the gait cycle.

[0068] S3. Based on the root mean square electromyography (RMS), calculate the mutual information (MI) between each target muscle and the exoskeleton auxiliary parameters, and determine the weight of each target muscle according to the mutual information MI, and construct a multi-muscle weighted cost function.

[0069] S4. Feedback the calculation results of the multi-muscle weighted cost function to the control end of the foot drop exoskeleton, and iteratively optimize the exoskeleton auxiliary parameters through the Bayesian optimization algorithm until convergence, thereby obtaining the optimal exoskeleton auxiliary parameters adapted to the user and completing the gait optimization of the foot drop exoskeleton.

[0070] In an optional embodiment of this application, the surface electromyography (EMG) sensor mentioned in step S1 is a wireless surface EMG sensor (Noraxon, Scottsdale, AZ, USA), and the sampling rate can be set to 2000Hz. The EMG signal collected by the surface EMG sensor is transmitted to the computer via a serial port. The computer aligns the EMG signal with the foot drop exoskeleton master control terminal through a socket synchronization flag to ensure that the EMG signal is synchronized with the time of the exoskeleton-assisted movement.

[0071] After initial assistance with the exoskeleton, electrical signals from the surface of muscles involved in dorsiflexion and plantarflexion of the ankle were collected. Ankle movement depends on the coordinated action of multiple muscles, such as... Figure 2 As shown, during dorsiflexion, the tibialis anterior (TA) acts as the prime mover, providing the primary dorsiflexion force; the soleus (SOL) and gastrocnemius (GAS) act as the primary antagonists, inhibiting dorsiflexion and driving plantarflexion. The soleus has a higher proportion of slow-twitch muscle fibers, providing the primary plantarflexion force during walking, while the gastrocnemius has a higher proportion of fast-twitch muscle fibers, participating more in high-intensity activities such as running and jumping, and playing a relatively smaller role in walking. With active exoskeleton assistance, the synergistic pattern of the dorsiflexion and plantarflexion muscle groups changes significantly; therefore, this invention selects to collect data from the TA, SOL, and GAS muscles. By analyzing muscle synergy to adjust the assistance, a safer and more comfortable experience can be provided for exoskeleton wearers.

[0072] In an optional embodiment of this application, the preprocessing in step S2 includes the following sub-steps:

[0073] S21. The heel-to-ground event is detected by the inertial measurement unit (IMU) at the toe of the foot drop exoskeleton, and the multi-channel surface electromyography signal is divided according to the gait cycle based on the heel-to-ground event.

[0074] S22. The segmented electromyographic signals are sequentially subjected to high-pass filtering, full-wave rectification, and low-pass filtering. The high-pass filtering is a second-order Butterworth filter with a cutoff frequency of 20Hz, and the low-pass filtering is a second-order Butterworth filter with a cutoff frequency of 10Hz. This processing can filter out most of the noise in the original signal and obtain envelope information about muscle activation.

[0075] S23. Calculate the root mean square (RMS) of electromyography (EMG) signals after filtering. In the process of calculating the RMS, the original EMG data in each gait cycle is uniformly resampled to several sampling points, and the RMS is calculated based on the amplitude of the several sampling points.

[0076] Specifically, the RMS calculation method is as follows:

[0077]

[0078] in This represents the total number of sampling points within a complete gait cycle. In this invention, the original electromyography data for each step needs to be uniformly resampled to 2000 points, therefore m ;also These represent the original electromyographic amplitude values ​​at points 1, 2, ..., m, respectively.

[0079] The aforementioned electromyography (EMG) signal acquisition time is determined based on a stability analysis method. This is because, to ensure that the data is not affected by factors such as initial adaptation and device drift, the subject typically needs to walk continuously for several minutes (2-5 minutes) to acquire a stable EMG signal. However, this is time-consuming for exoskeleton human-computer interaction feedback and has become a significant factor limiting optimization efficiency. Therefore, this invention estimates the minimum number of consecutive acquisition steps required for the EMG signal to reach a stable state, thereby obtaining representative muscle activity performance after discarding unstable data from the early stages, providing a basis for subsequent Bayesian optimization. For continuously acquired EMG signals, this invention proposes a signal stability judgment method as follows:

[0080] S201. Calculate the average RMS of the electromyographic signals acquired continuously from step 1 to step i (e.g., 2≤i≤100), denoted as A. i (i≥1), where A i This is the arithmetic mean of the electromyographic RMS from step 1 to step i;

[0081]

[0082] in, Representing the Step-by-step electromyography (RMS) Represents from step 1 to step 2 The average RMS of the step.

[0083] S202. Calculate the rate of change of adjacent average RMS using the forward difference method. All ΔA from step 1 to step i i Combined into a rate of change vector ;

[0084]

[0085]

[0086] S203. For all continuously acquired electromyographic signals, let l represent the total number of consecutive acquisition steps (l>1), then l-1 vectors a can be obtained. These vectors a are then divided into multiple sliding windows using a sampling rate of 1Hz and a window size of s=5. , where n is the window number;

[0087]

[0088] in, This represents the nth window. Next, to identify the steady state, a two-sample z-test is used to compare the average rate of change within two adjacent sliding windows.

[0089] S204, For adjacent sliding windows B n With B n-1 Perform a two-sample z-test, with the hypothesis being:

[0090]

[0091] in, The mean of the data within the window; specifically, the formula for the two-sample z-test is as follows:

[0092]

[0093] In the formula, Represents the nth sliding window Internal data mean represent Internal data standard deviation represent Internal data capacity, The representative test statistic can be used to calculate the two-tailed test. value:

[0094]

[0095] in, This represents the cumulative distribution function of the standard normal distribution. When the p-value of the two-tailed test is ≥0.95 (confidence level...),... When the multi-channel surface electromyography signal reaches a stable state, the acquisition is stopped and step S21 is executed.

[0096] Finally, after exploring 33 fixed exoskeleton-assisted modes (32 different assisted modes and 1 unassisted mode, each mode was completed by the same subject walking 100 steps continuously on a treadmill at a speed of 1.25 m / s (approximately 2 minutes),... This invention calculates the RMS value of electromyography signals for each step and normalizes it using the average maximum muscle activation value in the unassisted mode as the standard. The final conclusion is as follows: In different assisted modes, the average number of steps required for RMS stabilization is 55 steps (approximately 1 minute). This means that to accurately measure muscle performance under exoskeleton assistance, at least the data from the first minute must be discarded. Therefore, this invention clarifies that each user must walk continuously for at least 2 minutes. In addition, in the unassisted state, the average number of steps required for stabilization is 35 steps (approximately 40 seconds). This indicates that applying exoskeleton assistance may make muscle activity less stable when walking on flat ground. It is preliminarily believed that this may be related to factors such as exoskeleton comfort and leg symmetry.

[0097] In an optional embodiment of this application, the process of calculating mutual information MI in step S3 includes:

[0098] The root mean square (RMS) of stable electromyography (EMG) data acquired under various preset exoskeleton-assisted modes is calculated using arithmetic mean. The joint probability density of the averaged RMS and the corresponding exoskeleton-assisted parameters is then used. and marginal probability density , Calculate mutual information MI; where The average RMS is... These are auxiliary parameters for the exoskeleton.

[0099] It should be understood that mutual information (MI) is a metric in information theory used to measure the interdependence between two random variables, that is, how much information one variable contains about another. It is non-directional. It makes no specific assumptions about the distribution of variables or the type of relationship, making it suitable for analyzing unknown relationships between muscle activity with unknown distributions and exoskeleton auxiliary parameters. The formula for calculating mutual information is as follows:

[0100]

[0101] in, and These are two random variables whose correlation needs to be measured. In this invention, they represent muscle activity as expressed by RMS and a certain auxiliary parameter. yes and The joint probability density, and They are X and The marginal probability density is calculated using non-negative mutual information values. Since probability density estimation requires a large sample size (over 100), all pre-collected, continuously stable samples should be included in the estimation. However, even with overall stable performance, muscle activity varies between different steps. This variability may be caused by factors such as gait inconsistency and electromyographic signal noise. Considering this variability, probability density estimation will struggle to capture the RMS distribution differences between patterns. Therefore, in one embodiment, this invention averages the RMS data collected from all consecutive steps in each of the 33 patterns, aiming to amplify the RMS distribution differences between different assistance patterns through smoothing, allowing MI to more accurately reflect the dependence between the average muscle activation level and the exoskeleton assistance parameters. Nevertheless, it should be noted that due to the small sample size in this embodiment, the calculated MI may not accurately reflect this dependence; it can only be used to compare the differences in the dependence between different muscles and exoskeleton assistance to calculate relative weights.

[0102] In an optional embodiment of this application, the multi-muscle weighted cost function is:

[0103]

[0104] in As a cost, it is used to provide feedback on human-computer interaction performance to the exoskeleton. TA is the average RMS of the tibialis anterior muscle, SOL is the average RMS of the soleus muscle, and GAS is the average RMS of the gastrocnemius muscle. TA, SOL, and GAS are all normalized using the average maximum muscle activation value in the unassisted mode. , , These are the weights corresponding to muscle activity. Since mutual information is always non-negative, the absolute values ​​of the weights can be calculated by normalizing the mutual information to a range of 0 to 1.

[0105]

[0106] in This represents the mutual information between corresponding muscle activity and a certain auxiliary parameter of the exoskeleton. Since both SOL and GAS actively participate in ankle plantar flexion while resisting dorsiflexion, they are usually... and It was suggested that these values ​​be set to negative to maximize the difference between assists. However, an experiment in this invention found that such a design led to optimization favoring increased SOL and GAS activity, which often represents inappropriate ankle dorsiflexion assist. Therefore, this invention constrains all three weights to positive values ​​to balance the reduction of TA with the increase of SOL and GAS.

[0107] In an optional embodiment of this application, reference is made to Figure 3 , Figure 3 This paper presents a Bayesian optimization framework based on a weighted cost function, where muscle activity is estimated by the aforementioned weighted cost function and used to compute the posterior distribution of muscle activity with respect to gait. The posterior estimate is initially generated by evaluating six initial gait states. Given the posterior probability at the current iteration, the most probable auxiliary parameter is selected and applied to the exoskeleton. This process is repeated until convergence.

[0108] The Bayesian optimization algorithm described in step S4 includes the following sub-steps:

[0109] S41. Initialize the posterior distribution: Randomly select multiple predefined exoskeleton auxiliary parameters from uniform intervals, evaluate the multi-muscle weighted cost function under each parameter, and construct the initial posterior distribution of the objective function based on the evaluation results of each group.

[0110] Specifically, the initial posterior distribution of the objective function was obtained by evaluating the multi-muscle activation cost under six predefined auxiliary parameters. These parameters were randomly selected from uniformly spaced intervals and consist of two dimensions: peak force and peak time, as detailed in [link to documentation]. Figure 4 Gait is modeled as a bimodal curve, defined by mimicking human joint forces, enabling ankle dorsiflexion assistance during both the load response and swing phases. Peak force represents the maximum assist force provided by the exoskeleton, normalized to body weight, and is expressed in N / kg. Peak time represents the duration for which the exoskeleton provides maximum assist force, normalized to the gait cycle, and is expressed as a percentage.

[0111] S42. Constructing a surrogate model: Using a Gaussian process (GP) as a surrogate model to model the distribution of the objective function:

[0112]

[0113] Among them, the function The objective function is used to capture the relationship between exoskeleton auxiliary parameters and muscle activation during walking. It is the mean function. It is the covariance function. The covariance function of the Gaussian process is a combination of the Matern kernel and the White kernel with a smoothness parameter of 2.5. The Matern kernel is used to capture the smooth underlying structure of the objective function, and the White kernel is used to model the observation noise, which is the unstructured Gaussian noise in the electromyography signal acquisition process.

[0114] S43. Determine the next sampling point: Using the expected improved EI as the acquisition function, calculate the EI value of the candidate auxiliary parameters. The EI value is calculated based on the predicted mean μ(x) and predicted standard deviation σ(x) of the Gaussian process output. The next evaluation point is determined by maximizing the EI value. Furthermore, multiple random restarts are performed throughout the search space to avoid getting trapped in local optima;

[0115] In this embodiment, after modeling the posterior distribution of the objective function using GP, the acquisition function is used to determine the next sample point. This invention selects Expected Improvement (EI), which quantifies the expected improvement relative to the current best observation. Let... This is the best observation to date. At the candidate point... At that location, the GP will provide The predicted distribution, whose mean is The standard deviation is EI is defined as:

[0116]

[0117]

[0118] Where parameters It is a scaling factor used to balance exploration and utilization. and Let represent the standard normal cumulative distribution function and the probability density function, respectively. EI encourages sampling at locations with low predicted mean (i.e., potentially better than the current optimum) and high uncertainty (i.e., potentially unexplored regions). In each iteration, the next evaluation point... Maximize Function determined:

[0119]

[0120] Furthermore, this invention employs 10 random restarts throughout the search space to avoid getting trapped in local optima. The iterative process continues until a predefined stopping condition is met.

[0121] S44, Iterative Optimization: Move to the next evaluation point The corresponding exoskeleton auxiliary parameters are applied to the foot drop exoskeleton. Steps S1-S3 are repeated to obtain a new multi-muscle weighted cost function result. The surrogate model is updated based on the new result, and step S43 is executed again to determine the new... The iteration continues until a preset stopping condition is met, at which point the iteration stops and the current iteration is output. As the optimal auxiliary parameters for the exoskeleton.

[0122] After obtaining the next assessment point Afterwards (including peak force and peak time), the auxiliary parameters are converted into an assist trajectory in units of gait cycles using cubic spline interpolation, such as... Figure 4As shown. The above steps are repeated until convergence. During this process, the configured assist trajectory is transmitted through the flexible ankle exoskeleton of the cordless drive system, achieving personalized gait assistance.

[0123] Based on the above method embodiments, this embodiment also provides a foot drop exoskeleton gait optimization device based on multi-muscle coordination, which can be used to implement the steps of the above method.

[0124] refer to Figure 5 The foot drop exoskeleton gait optimization device based on multi-muscle coordination includes:

[0125] The acquisition module 51 is used to acquire multi-channel surface electromyography (EMG) signals related to ankle dorsiflexion and plantar flexion of the user through surface EMG sensors after the foot drop exoskeleton applies initial ankle dorsiflexion assistance to the user. The multi-channel EMG signals include at least the EMG signals of the tibialis anterior, soleus, and gastrocnemius muscles.

[0126] Preprocessing module 52 is used to preprocess the multi-channel surface electromyography signals to obtain the root mean square (RMS) electromyography of each target muscle during the gait cycle.

[0127] The cost construction module 53 is used to calculate the mutual information (MI) between each target muscle and the exoskeleton auxiliary parameters based on the root mean square (RMS) electromyography, and to determine the weight of each target muscle according to the mutual information (MI) to construct a multi-muscle weighted cost function.

[0128] The optimization module 54 is used to feed back the calculation results of the multi-muscle weighted cost function to the control end of the foot drop exoskeleton, and iteratively optimize the auxiliary parameters of the exoskeleton through the Bayesian optimization algorithm until convergence, so as to obtain the optimal auxiliary parameters of the exoskeleton that are suitable for the user and complete the gait optimization of the foot drop exoskeleton.

[0129] Various variations and specific examples of the methods provided in the above embodiments are also applicable to the apparatus of this embodiment. Through the foregoing detailed description of the methods, those skilled in the art can clearly understand the implementation method of the apparatus in this embodiment. For the sake of brevity, they will not be described in detail here.

[0130] To better execute the above-described method, embodiments of this application also provide a computer device for a foot drop exoskeleton, such as... Figure 6 As shown, the computer device includes a processor 61, a memory 62, and a communication bus 63 for enabling communication between the processor 61 and the memory 62.

[0131] Computer devices can be implemented in various forms, as long as they can realize the calculation process of this method.

[0132] The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the methods provided in the above embodiments; the data storage area may store data involved in the methods provided in the above embodiments.

[0133] Optionally, the memory may be a read-only memory (ROM), random access memory (RAM), electrically erasable programmable read-only memory (EEPROM), optical disc (including compact disc read-only memory (CD-ROM), compressed optical disc, laser disc, digital versatile optical disc, Blu-ray disc, etc.), magnetic disk storage medium, or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory exists independently and is connected to the processor via a communication bus, or the memory is integrated with the processor.

[0134] A processor may include one or more processing cores. The processor executes instructions, programs, code sets, or instruction sets stored in memory, and calls data stored in memory to perform various functions and process data as described in this application. The processor may be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller, and microprocessor. It is understood that, for different devices, the electronic devices used to implement the above-described processor functions may also be other types, and the embodiments of this application do not specifically limit this.

[0135] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0136] In an alternative embodiment, the computer device may also include a communication interface (not shown) for communication with other devices.

[0137] This application provides a computer-readable storage medium, including, for example, various media capable of storing program code such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. This computer-readable storage medium stores a computer program that can be loaded by a processor and execute the methods of the above embodiments.

[0138] This application also provides a computer program product comprising a computer program tangibly embodied on a readable medium thereof, the computer program containing program code for performing any of the methods described in any of the embodiments of this application, the computer program being downloadable and installable over a network, and / or installed from a removable medium (such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc.).

[0139] The above description of the embodiments is only used to provide a detailed introduction to the technical solutions of this application. However, the description of the above embodiments is only for the purpose of helping to understand the methods and core ideas of this application, and should not be construed as a limitation of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A computer device for a foot drop exoskeleton, characterized by, The processor, the memory and the communication bus are included, the communication bus is used for realizing the communication connection between the processor and the memory, the processor is used for executing the computer program stored in the memory, to realize the following multi-muscle coordinated foot drop exoskeleton gait optimization method, the method comprises: S1, after the foot drop exoskeleton applies initial ankle dorsiflexion assistance to the user, the surface electromyography sensor is used to collect the multi-channel surface electromyography signal related to the ankle dorsiflexion and plantarflexion of the user, and the multi-channel surface electromyography signal at least contains the electromyography signals of tibialis anterior muscle, soleus muscle and gastrocnemius muscle; S2, the multi-channel surface electromyography signal is preprocessed, and the root mean square RMS of the electromyography of each target muscle in the gait cycle is obtained; S3, based on the root mean square RMS of the electromyography, the mutual information MI between each target muscle and the exoskeleton assistance parameter is calculated, and the weight of each target muscle is determined according to the mutual information MI, and a multi-muscle weighted cost function is constructed; S4, the calculation result of the multi-muscle weighted cost function is fed back to the control end of the foot drop exoskeleton, the exoskeleton assistance parameter is iteratively optimized through the Bayesian optimization algorithm, until convergence, the optimal exoskeleton assistance parameter suitable for the user is obtained, and the gait optimization of the foot drop exoskeleton is completed.

2. The computer device for a drop foot exoskeleton of claim 1, wherein, The preprocessing in step S2 comprises the following substeps: S21, the inertial measurement unit IMU at the toe of the foot drop exoskeleton is used to detect the heel touch event, and the multi-channel surface electromyography signal is divided according to the gait cycle according to the heel touch event; S22, the divided electromyography signal is sequentially subjected to high-pass filtering, full-wave rectification and low-pass filtering, the high-pass filtering is second-order Butterworth filtering with a cutoff frequency of 20Hz, and the low-pass filtering is second-order Butterworth filtering with a cutoff frequency of 10Hz; S23, the root mean square RMS of the electromyography is calculated for the filtered electromyography signal, and in the calculation process of the root mean square RMS of the electromyography, the original electromyography data in each gait cycle is uniformly resampled to a plurality of sampling points, and the root mean square RMS of the electromyography is calculated based on the amplitude of the plurality of sampling points.

3. The computer device for a drop foot exoskeleton of claim 2, wherein, Step S2 further comprises a substep S20 of judging the stability of the multi-channel surface electromyography signal, which specifically comprises: S201、calculate the average RMS of the continuously collected first step to i step electromyographic signals, denoted as A i (i≥1), wherein A i is the arithmetic mean of the first step to i step electromyographic RMS; S202. Calculate the rate of change ΔA of adjacent average RMS using the forward difference method. i =A i+1 -A i All ΔA from step 1 to step i i The combination forms a rate-of-change vector; S203. sliding window division is performed on the change rate vector at a sampling rate of 1 Hz and a window size s = 5 to obtain a plurality of sliding windows B n where n is the window number. S204, performing a double-sample z test on the adjacent sliding windows B n with B n-1 A double-sample z test is performed, with the hypothesis being H0: μ(B n )=μ(B n-1 ), where μ is the mean value of the data in the window; when the p value of the double-tailed test is ≥0.95, it is determined that the multi-channel surface electromyography signal has reached stability, and the collection is stopped and step S21 is performed.

4. The computer device for a drop foot exoskeleton of claim 1, wherein, The process of calculating the mutual information MI in step S3 comprises: The stable root mean square RMS of electromyography collected under a plurality of preset exoskeleton assistance modes is respectively taken as an arithmetic mean, the mutual information MI is calculated based on the joint probability density p(x,y) and the marginal probability density p(x),p(y) of the averaged RMS and the corresponding exoskeleton assistance parameter, wherein x is the averaged RMS, and y is the exoskeleton assistance parameter; The determining the weight of each target muscle includes: normalizing mutual information MI i of each target muscle to the interval of 0~1 to obtain weight W i =MI i / ΣMI i , where ΣMI i is the sum of mutual information of tibialis anterior muscle, soleus muscle and gastrocnemius muscle, and the weight w1 of the tibialis anterior muscle, the weight w2 of the soleus muscle and the weight w3 of the gastrocnemius muscle are all positive numbers. The multi-muscle weighted cost function is CF=w1×TA + w2×SOL +w3×GAS, wherein TA is the average RMS of tibialis anterior muscle, SOL is the average RMS of soleus muscle, and GAS is the average RMS of gastrocnemius muscle, and TA, SOL and GAS are all normalized by the average maximum muscle activation value under the non-assistance mode.

5. The computer device for a drop foot exoskeleton of claim 1, wherein, The Bayesian optimization algorithm in step S4 includes the following sub-steps: S41, initializing the posterior distribution: randomly selecting a plurality of pre-defined exoskeleton assistance parameters from a uniformly spaced interval, evaluating a multi-muscle weighted cost function under each parameter, and constructing an initial posterior distribution of the objective function based on the evaluation results of each group; S42, constructing a surrogate model: using Gaussian Process GP as a surrogate model to model the distribution of the objective function, the covariance function of the Gaussian Process is a combination of Matern kernel with smoothness parameter 2.5 and White kernel, wherein the Matern kernel is used to capture the smooth underlying structure of the objective function, and the White kernel is used to model the observation noise, which is the unstructured Gaussian noise in the process of collecting electromyographic signals; S43, determining the next sampling point: using the expected improvement EI as the acquisition function, calculating the EI value of the candidate auxiliary parameter, the calculation of the EI value is based on the predicted mean μ(x) and the predicted standard deviation σ(x) output by the Gaussian process, and the next evaluation point is determined by maximizing the EI value , and multiple random restarts are performed in the entire search space to avoid falling into local optimum; S44, iteration optimization: the next evaluation point The corresponding exoskeleton assistance parameter is applied to the foot drop exoskeleton, and steps S1-S3 are repeated to obtain a new multi-muscle weighted cost function result, the agent model is updated based on the new result, and step S43 is executed again to determine a new , until the preset stop condition is met, the iteration is stopped, and the current is output as the optimal exoskeleton assistance parameter.

6. The computer device for a drop foot exoskeleton of claim 5, wherein, The exoskeleton assistance parameter is a two-dimensional parameter, including peak force and peak time; the peak force is the maximum value of the ankle dorsiflexion assistance force provided by the exoskeleton; the peak time is the time when the maximum assistance force is provided.

7. The computer device for a drop foot exoskeleton of claim 6, wherein, In step S4, the optimal exoskeleton assistance parameter is converted into an assistance trajectory in units of gait cycle by a cubic spline interpolation technique, and the assistance trajectory is transmitted to the user's ankle joint through the cordless driving system of the ankle drop exoskeleton, realizing personalized gait assistance.

8. The computer device for a drop foot exoskeleton of claim 6, wherein, The surface electromyographic sensor in step S1 is a wireless surface electromyographic sensor with a sampling rate of 2000Hz; the electromyographic signals collected by the surface electromyographic sensor are transmitted to the computer end through the serial port, and the computer end aligns the signals through socket synchronization mark between the electromyographic signals and the master control end of the ankle drop exoskeleton, ensuring the time synchronization of the electromyographic signals and the exoskeleton assistance action.

9. A multi-muscle synergy based foot drop exoskeleton gait optimization device, characterized by, It comprises: The acquisition module is used to collect the multi-channel surface electromyographic signals related to the ankle dorsiflexion and plantarflexion of the user through the surface electromyographic sensor after the ankle drop exoskeleton applies initial ankle dorsiflexion assistance to the user, and the multi-channel surface electromyographic signals at least include the electromyographic signals of the tibialis anterior muscle, the soleus muscle and the gastrocnemius muscle; The preprocessing module is used to preprocess the multi-channel surface electromyographic signals to obtain the root mean square RMS of the electromyographic signals of each target muscle in the gait cycle; The cost construction module is used to calculate the mutual information MI between each target muscle and the exoskeleton assistance parameter based on the root mean square RMS of the electromyographic signals, and determine the weight of each target muscle according to the mutual information MI, and construct a multi-muscle weighted cost function; The optimization module is used to feed back the calculation results of the multi-muscle weighted cost function to the control end of the ankle drop exoskeleton, and iteratively optimize the exoskeleton assistance parameter through the Bayesian optimization algorithm until convergence, so as to obtain the optimal exoskeleton assistance parameter adapted to the user, and complete the gait optimization of the ankle drop exoskeleton.

Images