Deep rehabilitation gait learning for modeling knee joints of lower-limb exoskeleton

Lower-limb exoskeleton is widely used for assisting walk in rehabilitation field. One key problem for exoskeleton control is to model and predict the suitable gait trajectories of wearer. In this paper, we propose a Deep Rehabilitation Gait Learning (DRGL) for modeling the knee joints of lower-limb exoskeleton, which firstly leverage Long-Short Term Memory (LSTM) to learn the inherent spatial-temporal correlations of gait features. With DRGL, the abnormal knee joint trajectories can be predicted and corrected based on wearer's other joints. This learning based method avoids gait analysis by building complex kinematic and dynamic models for human body and exoskeleton. More importantly, the new recovery gait pattern is not only in accordance with the healthy walking gait, but also including wearer's own gait profile. To verify the effectiveness of DRGL, a new recovery gait is obtained from DRGL based on “pathological gait” which is obtained by a healthy subject imitating knee injury. Experiments demonstrate that the subject can walk normally with SIAT lower-limb exoskeleton in new recovery gait pattern.