An Integrated Virtual Hand Platform for Evaluation of Model-Based Control of Hand Prosthesis.

Virtual environments are commonly used in system validation and pre-wear training for prosthetic hand systems. In this study, we developed a human-like tendon-driven virtual hand system in MuJoCo emulation environment. The virtual hand was integrated with external neuromorphic models of flexor and extensor muscles acting on digit tendons. Contact pressures at prosthetic fingers were conveyed to an external sensory feedback system. The neuromorphic model emulated a monosynaptic neuromuscular reflex loop. A pair of $\alpha$ motor commands derived from EMG signals of amputee's arm activated neuromorphic muscles via motoneuron pools that were also regulated simultaneously by spiking afferents from muscle spindles. Using the integrated virtual hand system, we showed that control of equilibrium positions (EPs) of fingertip can be achieved by co-contracting $\alpha$ motor commands of antagonist neuromorphic muscles. Responses to force perturbations applied to the fingertip during maintenance of EP allowed calculation of stiffness ellipses at the fingertip. The biomimetic controller could switch automatically between position and force control, or stiffness regulation during stable grasping of a spring with different stiffness. It was demonstrated that the virtual hand platform is useful for amputees to explore biomimetic neuromuscular control of hand grasp with real-time tactile sensory feedback.