Curious, could one apply fancy control theory and modern computing at the problem and fix most of the accuracy and flex problem? Position feedback with integrated sensors can be very compact and highly accurate, and human muscle seem to also use feedback control for holding position..