Predictive Locomotion Mode Recognition and Accurate Gait Phase Estimation for Hip Exoskeleton on Various Terrains

In recent years, lower-limb exoskeletons have been applied to assist people with weak mobility in daily life, which requires enhanced adaptability to complex environments. To achieve a smooth transition between different assistive strategies and provide proper assistance at the desired timing during locomotion on various terrains, two significant issues should be addressed: the delay of locomotion mode recognition (LMR) and the accuracy of gait phase estimation (GPE), which are yet critical challenges for exoskeleton controls. To tackle these challenges, a high-level exoskeleton control, including a depth sensor-based LMR method and an adaptive oscillator-based GPE approach, is developed in this study for terrain-adaptive assistive walking. An experimental study was conducted to evaluate the effectiveness and usability of the proposed control in a real-world scenario. Experimental results suggested that the depth sensor-based LMR method can detect the locomotion mode change 0.5 step ahead of the assistive strategy switch of the leading leg, while the average environment classification accuracy across five subjects was higher than 98
$\%$
. The accuracy is comparable with the state-of-the-art LMR methods, but its predictive capability is beyond existing LMR methods applied in lower-limb exoskeletons. Moreover, the adaptive oscillator-based GPE approach accurately estimated the user’s gait phase during locomotion on various terrains, with a root-mean-square (RMS) gait phase reset error of only
$\pm$
0.27
$\%$
, outperforming the literature standard.