Enhancing Gait Trajectory Prediction through an Adaptive Template Selection Mechanism

Adaptive Template Switching for Efficient Gait Prediction

Abstract

In the field of human-robot interaction for exoskeletons, the demand for efficient and real-time monitoring of human gait is crucial. In this paper, we enhance the dynamic template clustering algorithm to boost the accuracy and stability of gait trajectory prediction. We introduce the local optimum rule and the optimal conservative rule for template switching, tested on a real-world gait dataset. Ideally, with all template performances known, we achieve an R² of 0.904. In embedded systems, where template performance is unknown, the local optimum rule runs five templates to select the lowest local window loss, achieving an R² of 0.880. The optimal conservative rule, favouring minimal template switching, attains an R² of 0.889 with introduced sleep time and threshold selection parameters. This approach balances computational demand and accuracy, significantly enhancing the efficiency and efficacy of exoskeleton systems and wearable rehabilitation devices.

DOI: https://doi.org/10.24135/ICONIP21