Robots will soon become a domestic commonplace but there true potential will be realized only with the perfection of contact robotics, machines that cooperate physically with humans.

Forceful human-machine interaction engenders surprising challenges. Even the deceptively simple act of pushing on a tool is statically de-stabilizing. Yet humans accomplish contact tasks with consummate ease, due, in part, to skillful control of mechanical impedance. I will show that the limits of human force production are determined by impedance rather than strength.

Endowing machines with similar characteristics may enable robots to approach (or exceed) biological performance, but how best to achieve controllable impedance remains a challenge, especially at forces comparable to human strength. Force feedback is appealing but contact imposes severe stability limits. I will review recent progress using quantitative knowledge of human operator impedance to reduce force feedback conservatism, an example of the synergy between motor neuroscience and engineering.

Feedback control may be augmented by engineering novel biomimetic actuators. Electro-active polymers may ultimately achieve muscle-like function but their application requires control-relevant models--minimally complex yet reproducing essential behavior. I will show how an energy-based network approach yields real-time computable models that accurately reproduce observed behavior over the full range of excitation and--more important--provide new insight.