Real-Time Gait Events Detection during Walking of Biped Model and Humanoid Robot through Adaptive Thresholds

Abstract

Temporal abnormalities in events and gait phases of human walking can be potentially applied in the gait performance analysis. In this study is presented a novel real-time gait events detector that continuously considers the previous gait motions in order to improve this gait analysis. The detection of heel-strike and toe-off events is performed by a finite state machine (FSM), through decision rules and adaptive thresholds. The proposed algorithm stands out of the state-of-the-art by using adaptive thresholds in the FSM's decision rules, making the proposed gait event detector robust to sporadic perturbations, and adaptive to locomotion mode changes. For such, three stages were considered: thresholds' calibration, real-time detection of gait events and thresholds' update. Anatomical differences between lower limbs demanded independent FSMs for each limb. The algorithm was validated in a simulated biped model, and on a real DARwIn-OP robot, what shows up the generality of the proposed approach. Results highlight that the proposed algorithm correctly detects the gait events (accuracy of 100% and 84.348% in simulated and real conditions, respectively), with average time delays from 15.5 to 34ms. Thus, the proposed detector is an adaptive, accurate and versatile tool for real-time analysis of the feet movements along gait cycle.