Neural oscillator network–based biped control for self-adapting gait

Abstract

Walking on floors with a varying slope needs more adaptive walking controller against the slope change, since that kind of walking becomes unstable easily without visual information. It may be difficult even for human beings to keep the walking stability without seeing the slope. This work presents a neural oscillator network to generate the patterns for periodic bipedal locomotion, which enable a humanoid robot to adapt to slope change of terrain. Motion trajectories of each limb (each hand and foot) are first defined in terms of periodic functions, the coefficients of which are the output parameters of neural oscillators. Those parameters are determined with the neural oscillator network in cooperation with sensory signals that detect the states of feet in contact with the terrain such that the motion trajectories are scaled for the walking stability. In addition, for the same reason, the neural oscillator controls the trajectories of the center of mass and the zero moment point of humanoid. Using the proposed method, the walking of the humanoid was performed on uneven and uncertain terrain. This application for the humanoid robot may draw some helpful hints on understanding human beings’ walking mechanism against the terrain with a varying slope.