Non-linear adaptive control inspired by neuromuscular systems

Current paradigms for neuromorphic computing focus on internal computing mechanisms, for instance using spiking-neuron models. In this study, we propose to exploit what is known about neuro-mechanical control, exploiting the mechanisms of neural ensembles and recruitment, combined with the use of second-order overdamped impulse responses corresponding to the mechanical twitches of muscle-fiber groups. Such systems may be used for controlling any analog process, by realizing three aspects: Timing, output quantity representation and wave-shape approximation. We present an electronic based model implementing a single motor unit for twitch generation. Such units can be used to construct random ensembles, separately for an agonist and antagonist 'muscle'. Adaptivity is realized by assuming a multi-state memristive system for determining time constants in the circuit. Using SPICE-based simulations, several control tasks were implemented which involved timing, amplitude and wave shape: The inverted pendulum task, the 'whack-a-mole' task and a handwriting simulation. The proposed model can be used for both electric-to-electronic as well as electric-to-mechanical tasks. In particular, the ensemble-based approach and local adaptivity may be of use in future multi-fiber polymer or multi-actuator pneumatic artificial muscles, allowing for robust control under varying conditions and fatigue, as is the case in biological muscles.