Abstract
This thesis deals with the problem of humanoid falling with a decoupled strategy consisting of a pre-impactand a post-impact stages. In the pre-impact stage, geometrical reasoning allows the robot to choose appropriateimpact points in the surrounding environment –that can be unstructured and may contain cluttered obstacles,and to adopt a posture to reach them while avoiding impact singularities and preparing for the post-impact. Thepost-impact stage uses a quadratic program controller that adapts on-line the joint proportional-derivative (PD)gains to make the robot compliant, i.e. to absorb post-impact dynamics, which lowers possible damage risks.We propose a new approach incorporating the stiffness and damping gains directly as decision variables in theQP along with the usually-considered variables that are the joint accelerations and contact forces. By doing so,various constraints can be added to the QP. Finally, since the gain adaptation is local, we added a preview ona time-horizon for more optimal gain adaptation based on model reduction. At each step of the development,several experiments on the humanoid robot HRP-4 in a full-dynamics simulator are presented and discussed.