Reconstruction of Hip Moments Through Constrained Shape Primitives

Abstract

Among the various control laws developed for lower-limb wearable robotics, several intend to replicate leg joint moments observed in healthy humans. While this can be achieved by different means, inspiration can be gained from the so-called motor primitives, i.e. a small set of fundamental signals used by the nervous system to recruit numerous motor pools in a task-specific way. Here we report a methodology for the extraction of constrained-shape primitives. Stimulations of two antagonist hip muscles are initially retrieved from a database composed of different locomotion tasks. A first guess of the primitives parameters is then performed in order to reconstitute these stimulations. Finally these parameters are optimized with the aim of retrieving the original database moments. Our findings suggest that a small number of symmetric Gaussian-like periodic primitives may reconstruct these hip moments with a level of fidelity that would be sufficient for providing a task-specific user support.