Towards a Generalized Model of Multivariable Ankle Impedance During Standing Based on the Lower Extremity Muscle EMG

The ankle mechanical impedance of healthy subjects was estimated during the standing pose while they co-contracted their lower-leg muscles. Subsequently, the impedance parameters were modeled as a function of the level of co-contraction using machine learning regression methods. From the experimental results, the average ankle stiffness coefficients in dorsi-plantar flexion (DP) showed more dependence to the muscle contraction than stiffness in inversion-eversion (IE): 4.6 Nm/rad per %MVC (percent of the maximum voluntary contraction) and 1.1 Nm/rad per %MVC, respectively. To accurately estimate the ankle impedance parameters as a function of the electromyography (EMG) signals, multiple EMG feature selection methods, regression models, and types of models were evaluated. Using a 1-vs-All model validation approach, the best regression model to fit the stiffness and damping in DP was the Least Square method with Regularization, and the best IE stiffness was the Gaussian Process Regression. No model was able to estimate the IE damping well, possibly because this parameter is not modulated with a changing co-contraction of the lower-leg muscles.