Gait generation for a biped robot with knees and torso via trajectory learning and state-transition estimation

Abstract

The proposed method can generate an optimal feedforward control input and the corresponding optimal walking trajectory minimizing the \(L_2\) norm of the control input by iteration of laboratory experiments. Since a general walking motion involves discontinuous velocity transitions caused by the collision with the ground, the proposed method consists of the combination of a trajectory learning part and an estimation part of the discontinuous state transition mapping using the stored experimental data. We apply the proposed method to a kneed biped robot with a torso, where we also provide a technique to generate an optimal gait not only being energy-efficient but also avoiding the foot-scuffing problem.