Iterative Learning of Human Behavior for Adaptive Gait Pattern Adjustment of a Powered Exoskeleton

Abstract

Powered exoskeletons for people with complete paraplegia have been controlled based on predefined joint-reference trajectories. As the target users of such robots may not realize any voluntary movement, the human body is fully constrained and follows the movement of the powered exoskeleton joints. The predefined gait pattern, however, may or may not be adequate for every user because the gait pattern is resulting from complex interactions between the body segments and environment, as well as dynamic characteristics of the body segments. As all the persons and their body segments have different dynamic characteristics, therefore, a bespoke tuning of gait parameters is necessary in order to realize the natural gait motion, which is optimal for each user. In this article, an adaptive gait pattern adjustment method is proposed. The proposed method observes the ground contact timing, which is directly related to the adequacy of the gait pattern for the user wearing a robot. Based on the ground contact timing, the joint-reference trajectories are adjusted, which are parameterized by the trunk inclination angle. The proposed method iteratively calculates an appropriate trunk inclination angle from the information of ground contact timing. In this article, the derivation of the proposed method and its experimental verification with WalkON Suit, a powered exoskeleton, are introduced. The proposed algorithm successfully worked for two practical users with complete paraplegia, and the adapted gait patterns showed excessive performance in walking speed, oxygen consumption, palm force on crutches, etc. The results were also verified by winning both gold and bronze medals in the global competition, Cybathlon 2020, while accomplishing the best records among all the teams.