Extraction of simple monophasic motor primitives towards bio-inspired locomotion assistance

Abstract

Advanced analyses of muscle activation signals during various locomotion tasks revealed that these large sets of profiles can be reconstructed by a much smaller amount of fundamental signals. These so-called “primitives” are thus putatively presented as the building blocks composing large and diverse movement repertoires. Previous results validated the use of such bio-inspired primitives for assisting locomotion through wearable robots relying on a simple and modular framework. Various locomotion tasks such as walking and stairs ascending/descending can be assisted by delivering torque profiles that are composed from this limited set of primitives. This paper goes one step further and presents a framework for extracting such primitives while constraining them to be simple and monophasic. More precisely, these primitives are shaped by three parameters only: their peak, their width, and a potentially asymmetry factor de-correlating the duration of the rising and falling edges. The paper shows that five primitives extracted from walking and running data might be enough to reconstruct virtual muscle stimulations across a large range of leg muscles and locomotion tasks, and that releasing primitive asymmetry only brings a limited improvement in the signal reconstruction process. This paves the way towards a versatile and extremely computationally efficient framework for locomotion assistance through wearable robots.