Falling Detection for Humanoid Robot based on MOA Set with Macroscopic Feedback Gain

Abstract

Motion control of a humanoid robot is challenging problem because its dynamics is complicated. Many studies employ a simple model focusing on a macroscopic dynamics between the center of gravity (COG) and the center of pressure (COP). This model makes it easier to plan a referential trajectory, design a controller, and analyze its performance. In particular, analysis on the macroscopic dynamics provides us with an index for falling detection. In previous studies, the authors proposed a falling detection method based on the maximal output admissible (MOA) set and its experimental computation method. We can compute the MOA set from macroscopic feedback gain which is identified from disturbance response in an actual robot. In this paper, the validity of this method is investigated more thoroughly by full-body dynamic simulations.