Continuous Gait Phase Estimation from Translational Kinematics: Towards Implementation in Powered Ankle Prostheses

Abstract

For people with transtibial amputations (PwTTA), discontinuities in powered prosthesis control schemes can lead to a disastrous trip or fall. Thus, continuous gait phase estimators have gained attention to facilitate smoother and more natural feeling device control. However, existing continuous estimators often struggle to generalize between steps, speeds, tasks, people, and impairments. This work 1) presents a novel method to continuously estimate the gait phase from multi-segmental, anterior-posterior translational kinematics that naturally aligns with key gait cycle events, 2) enhances this estimator through linear regression of within gait cycle features from an open-source dataset of people walking::; 1.2 m/s, and 3) validates the estimator's performance on two datasets independent from regression training. The first evaluation compared the regressed estimator against an existing adaptive oscillator approach on a dataset of young adults (YA) without impairment, walking between 0.4-1.6 m/s. Analysis revealed that our estimator performs more consistently between strides, gait speeds, and individuals. The second validation compared the YA results against a group of PwTTA walking with passive and powered ankle prosthesis between 1.0-1.2 m/s, finding no significant differences in gait phase estimation accuracy between groups or devices. These promising offline results are suggestive of future success for our gait phase estimator in robotic ankle prostheses and other applications.