Humanoid fall avoidance from randomly directed disturbances

Abstract

When humanoid robots are pushed they risk falling and damaging themselves. Preemptive fall avoidance strategies are becoming popular and usually involve measures such as actively actuating the ankles, the hips or even taking steps. Deciding which strategy to take can be determined based on a stability region—known as the decision surface—drawn on a phase plot of the robot's state. Unfortunately, the decision surface is limited to disturbances emanating from the sagittal or coronal planes. This paper addresses this limitation by proposing a decision hypersurface for a hip strategy, which is used for the prediction of limiting states for recovery from disturbances in any orientation. A second contribution is the extension of the hip strategies for humanoid fall avoidance to disturbances in random directions. Both, strategies and decision hypersurfaces are tested on the Webots simulator then implemented on a real humanoid robot.