Improving Force-Based Interaction Control Using Limited Knowledge of Environment Dynamics

Robots for neurological rehabilitation require actuators with high force/weight ratio and low mechanical impedance. Conventional actuators with high force capacity are generally either extremely heavy (e.g. electromagnetic actuators) or have high endpoint impedance (e.g. geared actuators). Measurement of the endpoint force can be used with feedback control to reduce the apparent impedance, but this can seriously compromise stability when the system couples to a dynamic environment such as a human limb. Energy-based controller design methods are typically used to guarantee stability, but the result is extremely conservative. In this work, knowledge of the bounds of the dynamic characteristics of human limbs (e.g. stiffness, mass, damping) is used with robust stability tools to shape dynamic force feedback compensators to reduce impedance more aggressively without sacrificing guaranteed stability. In the video clip, one result is shown on an actuator testbed. The system is capable of 120 N continuous force but has 15-20 N of nonlinear, position-dependent friction. With the new compensator the apparent friction is reduced to less than 0.5 N (and inertia is reduced by around 80%), such that the system can be backdriven with a potato chip, but coupled stability remains robust enough to stabilize impact with a hard plastic block.