Online learning of COM trajectory for humanoid robot locomotion

Abstract

Center Of Mass (COM) trajectory is an essential factor for stable and natural robot locomotion. Unlike previous research in which COM trajectory either restricted by ZMP trajectory or directly predefined by simple function such as sinusoid, this research aims to establish the COM trajectory by online autonomous learning under the objective of locomotion stability and naturalness, which is expressed as a self-consistent measure in this paper. It provides an alternative that may avoid or weaken the mismatch between theoretical planning and practical implementation. The experimental results on a real humanoid robot PKU-HR4 show its effectiveness and promising future.