A Study on Stair Gait Phase Detection and Stair-Aware Module for Lower-Limb Exoskeletons

This thesis aims to improve lower-limb exoskeletons that assist individuals with mobility impairments, especially during stair walking. Unlike level-ground walking, stairs introduce unique biomechanical challenges. The research proposes two supervised learning models: an LSTM-CRF model for detecting the user’s gait phase using IMU sensors, and a ViG (Vision-based Graph Neural Network) model for recognising locomotion intention using camera input. Both models were trained on real-world datasets. Results demonstrate enhanced accuracy and reliability across different classes, making exoskeletons more responsive and better suited for use in everyday, unstructured environments beyond controlled laboratory settings.