A phase-shifting based human gait phase estimation for powered transfemoral prostheses

Abstract

Phase variables are continuous real-valued functions which parameterize the gait cycle. As such, they can be used to estimate a powered prosthesis user’s walking state. It is common to represent the phase variable as a function of the thigh angle. This approach assumes the thigh angle profile and its integral to be sinusoids. However, this assumption does not hold at lower walking speeds which leads to inaccurate heel-strike detection. In this study, we show that the thigh angle and its integral are phase-shifted from the ideal sinusoids by analyzing both able-bodied and amputee walking data. We then propose a novel phase variable incorporating phase-shift that is accurate even at lower walking speeds. We tested two variants of the proposed phase-shifted phase variable on a transfemoral prosthesis in an emulator study. Results show that phase-shifting improves heel-strike detection. Additionally, phase-shifting improves the linearity of the phase variable over the gait cycle. Analysis of the knee and ankle phase portraits showed that phase-shifting results in fewer deviations from the limit cycle.