Unit step walking pattern generation for a humanoid robot

Abstract

In this thesis, an analytic method for generating a unit step pattern using a center of mass (COM) constraint is presented. The walking pattern is based on the linear inverted pendulum model (LIPM). Assuming radical zero moment point (ZMP) trajectories, a simple analytic solution form of the COM trajectories is formulated. Four types of case studies are presented for analytic solution, based on different constraints on ZMP references, COM, and time differences. The unknown parameters of the COM trajectories in the solution form are formulated for the different cases. Using the unit step pattern sequence, various types of walking pattern are generated. In case study, fourth case , “ COM first and symmetrical ZMP later with time constraint”, is ultimately propose method in this thesis. The proposed method was verified by experiment of long stride walking with the DRC-HUBO robot developed at KAIST for DRC (DARPA Robotics Challenge). Experiments of long stride walking on bricks were successfully performed using the unit step pattern with several previous HUBO’s controllers. In addition, visual assist supervisory walking strategies are proposed for the DRC trial 2013. These new strategies are achieved using unit step walking pattern.