Estimation of Three-directional Ground Reaction Force during Walking Using Sole Sensor System

Abstract

There is a demand for the development of a ground reaction force measurement system that is not subject to location restrictions. In the current study, we developed a sole sensor system equipped with four high-capacity compact 3-axis force sensors using a Cr-N thin film. Using the sole sensor system, time-series data of the three-directional ground reaction forces during straight walking was estimated. Two models, a multiple regression analysis model and a Gaussian process regression model, were used for the estimation. The % root mean square error (%RMSE) and R² values in the multiple regression analysis were 14.3 % and 0.419 in average in the left-and-right directions, respectively. The %RMSE and R² values in the anteroposterior direction were 8.0 % and 0.869 in average, respectively, and those in the vertical direction were 8.2 % and 0.801 in average, respectively. The %RMSE and R² values in the Gaussian process regression were 13.6 % and 0.458 in average in the left-and-right directions, respectively. The %RMSE and R² values in the anteroposterior direction were 6.5 % and 0.891 in average, respectively, and those in the vertical direction were 4.8 % and 0.910 in average, respectively. In addition, in both models, the estimation accuracy in the 0-20% stance phase, which corresponds to the heel sensor contacting period, was lower than that in other stance phases.