Control of Humanoid Robots

Abstract

Since “Asimo” was introduced by Honda in 2000, humanoid robots have drawn increasing attentions of both academics and general public. To make a humanoid robot function as expected, control plays a vital role. Unfortunately, due to the difficulties in implementing a full-dynamics-based real-time control for a humanoid robot with hyper degrees of freedom, most state-of-the-art control approaches use ZMP (zero moment point) technique in which the dynamic interactions among different parts of a robot are treated as disturbances and only the gravity compensation with a rough dynamic force prediction is integrated into the decentralized joint PID controller. The lack of a precise modeling, without doubting, limits the control performances of biped systems and consequently limits the speeds in dynamic walking.