Advancing Mobility in Stair Climbing with BART LAB's Intelligent Wheelchair: A Deep Learning Approach to Pose Estimation

This study addresses the critical urban mobility challenges faced by individuals with lower-limb impairments, particularly the limitations of conventional wheelchairs in navigating stairs of varying dimensions. Conventional wheelchairs struggle with these challenges, particularly those with non-uniform stair dimensions, which impede urban accessibility. To overcome these limitations, we introduce a semi-autonomous tracked stair-climbing wheelchair (SCW) that features an innovative kinematic mechanism that seamlessly switches locomotion modes and autonomously adjusts the wheelchair's pose to adapt to diverse terrains. The research focuses on enhancing pose estimation capabilities during stair climbing through a novel deep-learning approach. By integrating a depth camera with an advanced kinematic model, a sophisticated algorithm capable of precise stair width estimation and pose adjustment has been developed. Experimental results confirm the system's effectiveness, achieving a Mean Absolute Error (MAE) of 0.05 meters and a Root Mean Squared Error (RMSE) of 0.07 meters in height measurement, with high precision (0.92) and recall (0.89) rates in identifying stair components. These findings demonstrate the system's capability to overcome accessibility and safety challenges, offering a technical stride towards improved inclusivity and independence in urban environments.