Phase Learning to Extract Phase from Forelimb(s) and Hindlimb(s) Movement in Real Time

Abstract

Interlimb coordination is important for the enhancement of walking gait in spinal cord injured patients and many studies have recently attempted to dynamically map these movements for use in assistive devices. Nevertheless, there are many difficulties such as high variation of signal and lack of precise algorithms to extract continuous phases in real time. An improved phase learning to extract forelimb(s) and hindlimb phases from movements in real time is proposed. To quantify the performance of our proposed phase learning method, this phase learning is compared to Hilbert transform, a commonly used analytical method for offline process, with principal component analysis (PCA). The comparison between two methods demonstrated that a percentage of root mean square (RMS) time error between goal phase and output phase from our phase learning method is 7.94% as compared to that of Hilbert transform (7.44%). This phase learning that can extract phase in real time improves the analysis of interlimb coordination in robotic application.