Embodied interaction control from human to artificial muscles

Abstract Soft actuators have been extensively developed over the past two decades, yet their control strategies remain rudimentary and do not exploit well their unique viscoelastic properties. To develop skilful control of soft actuators, we take inspiration from human sensorimotor control, which achieves dynamic and accurate movements despite relying on noisy and slow muscles. We first examine how the human nervous system (HNS) optimally controls muscles to exchange energy with the environment and extract maximal information from it. Critically, the HNS prepares interactions by learning specific patterns of reciprocal activation and co-activation, thereby regulating force and impedance, storing elastic energy and embodying uncertainty. We then show that soft actuators share key mechanical characteristics with human muscles and could thus benefit from recently identified computational mechanisms of the HNS, yielding efficient nonlinear adaptive impedance and stochastic nonlinear optimal control algorithms.