Developing a Human-Machine Control Interface for the Detection of Motion Intentions in a Self-Balanced Lower-Limb Exoskeleton

Abstract

In the recent years, advancements in robotics-related fields accompanied the development of exoskeletal devices that enhance the physical capabilities of the wearer, or assist impaired individuals in performing specific body movements. In particular, assistive lower-limb exoskeletons can be proposed to impaired people as a possible alternative to wheelchairs, or as rehabilitation medical devices. However, the intention detection interfaces are often based on basic solutions that lack intuitiveness, partly monopolize the use of hands, or prevent seamless transitions between the available activity modes. In this context, this doctoral work investigates natural and intuitive movement- based solutions to robustly detect motion intentions in a marketed assistive lower-limb exoskeleton. It focuses on walking-related intentions – namely gait initiation, gait termination, and steering – and evaluates novel implementations of high-level controllers based on acceleration and angular velocity signals recorded from upper-body-worn Inertial Measurement Units. Signals from these sensors can be analyzed, so that descriptive features of the exhibited movements are extracted, and serve as inputs to a classification architecture: they can either be compared to training data in a supervised learning approach, or to empirically derived thresholds. Experimental results of these algorithms indicate that the developed methods could be a viable alternative for intention detection in medical lower-limb exoskeletons, and could greatly enhance their usability.