(A) study on analytic method of zero-moment point pattern generation of humanoid robot based on capture point

Abstract

To enable a humanoid robot to walk stably without falling down, basic walking patterns as well as walking controllers are needed. We introduce analytic solution methods to generate two kind of safe feasible walking patterns based on a capture point (CP)

in other words, a divergent component of motion or an extrapolated center of mass. One method is that use a continuous zero-moment point (ZMP) pattern with linear-constant functions and the other method is that use an optimal ZMP pattern. And we define $h$ value for convenience of analysis. The $h$ value is a convolution term of analytic solution of equation of motion and the target CP is determined by this $h$ value. In all two kind of methods, $h$ value, which need to reach a target CP, is determined and from this $h$ value two kind of continuous ZMP patterns are generated within convex hull of the support feet area. We compare and analyze these two kind of generated walking patterns and experimentally compare energy efficiency from the measured power consumption when the humanoid robot walks with those walking patterns. And we extend the method to generate a real-time walking pattern such as step length change while walking and push recovery during walking on the spot. Furthermore, we introduce a ZMP compensator to control the humanoid robot and a double and single inverted pendulum model with flexible joints, which is used in the ZMP compensator. Through the model-based control, we resolve oscillation from compliant characteristics of the humanoid robot. Finally, we experimentally verify two kind of walking pattern, the real-time walking pattern generation method, and the walking control scheme with the humanoid robot, HUBO2.