COMPLIANT CONTROL OF ELASTIC ACTUATORS FOR HUMAN ROBOT INTERACTION

This thesis focuses on the control of interaction between a soft robot and a human being. It proposes a novel approach to physical human-robot interaction (pHRI) by explicitly accounting for human dynamics. In fact existing interaction control solutions guarantees stability by regarding the human to a passive system without accounting for its dynamics. Unfortunately such unmodeled dynamics is actually in the loop and influences the interaction unpredictably. As a result the closed loop dynamics is stable but not well defined. This is in general not desired especially if we aim to guarantee a certain control performance as in the case of robotic orthoses and prostheses. Also unpredictable interaction dynamics can be harmful when closing an higher level control loop as the stability region is in turn not well defined. This thesis proposes a novel control approach to the problem. We start from the analysis of existing design and control principles of soft interaction. We outline that physical compliance has the fundamental role of stabilizing force control. Then we propose a novel approach that, taking advantages of compliance, explicitly accounts for the human dynamics in the loop. In particular we propose two kind of control solutions. The first is based on the on-line estimation of human dynamics and on-line adaptation of control law. The second considers certain unmeasurable parts of human dynamics as a disturbance and provides to be insensitive to it. Stability and performance of both solutions are theoretically guaranteed under realistic hypotheses. Experimental validation in a physical human-robot interaction task shows evident advantages of the proposed approaches with respect to existing solutions.