Walking Pattern Generation using Approximate Unstable Zero Cancelation and Its Compensation Method

Abstract

This paper suggests a walking pattern generation method using an approximate unstable zero cancelation and its compensation. Basically, there exists an error in the walking pattern generation method for humanoid robots using both a pole-zero cancelation by series approximation (PZCSA) and a linear quadratic regulator (LQR) method suggested in [12]. This error is caused by the approximate pole-zero cancelation for removing an unstable zero. As an alternative of this problem, firstly, we construct an estimator for position/acceleration errors of Center of Mass (CoM) and a position of Zero Moment Point (ZMP), and then they are used for compensating the ZMP error and the CoM error using a low pass filter. Through these simple procedures, finally, we show that both the ZMP and CoM errors can be removed through numerical simulations.