Walking pattern prediction with partial observation for partial walking assistance by using an exoskeleton system

Abstract

Movement prediction is a key ingredient in exoskeleton robot control for walking assistance. In this paper, we propose a movement prediction method with following two desirable fundamental properties: 1) fast online calibration for a novel user, and 2) applicability to partially observable situations. Using this method, for example, 1) we can use previously collected other subjects' walking data to quickly adapt to a novel user's movements in exoskeleton robot control, or 2) we can generate the exoskeleton robot movement for assisting right leg behavior by only observing the movement of the left leg. To validate our proposed method, we conducted experiments in walking movement prediction using a one-leg three DOFs exoskeleton robot with nine healthy subjects. The experimental results suggest that our method is able to predict a new user's walking pattern and to cope with the partial observations.