Humanoid fall avoidance from random disturbances predicted via a decision volume

Abstract

Humanoid fall avoidance is the ability of a robot to avoid falling when pushed. The decision surface is a region on the phase diagram delimiting the states beyond which the robot cannot recover from a disturbance. The disadvantage of the decision surface is that it is limited to perturbations in the sagittal and coronal planes. This paper deals with the generalization of the decision surface to a decision volume, used for the prediction of limiting states for recovery from disturbances in any orientation. A second contribution is the extension of the ankle strategy for humanoid fall avoidance to disturbances in random directions. The model used is a 3D Linear Inverted Pendulum Model (LIPM). Both, ankle strategy and decision volume are tested on the Webots simulator then implemented on a real humanoid robot.