Walking control of humanoid robot based on extreme learning machine

This paper investigates the dynamic balance problem of humanoid robot and presents a systematic control architecture. In order to achieve better locomotion stability and control performance, a hybrid offline and online control algorithm based on all robot joints is proposed. Considering the complicated nonlinear relationship between zero-moment-point (ZMP) and robot joints, an offline learning algorithm based on extreme learning machine (ELM) is adopted to approximate the centre of mass (CoM) correction value according to ZMP error. Then, an online control method is employed to adjust all joints trajectories according to CoM position by minimising energy consumption. Given the optimised joints motion, an adaptive control system is proposed to track the desired trajectories and the stability proof is provided. The simulation results validate the proposed method.