Abstract
In this paper, we propose a method to generate various walking gait patterns for biped robots in real-time using an approximated dynamics model with two point masses representing motion of the feet and a linear inverted pendulum representing the horizontal motion of the torso. The motion of the inverted pendulum is calculated from the trajectories of the feet and the desired ZMP. Supposing these trajectories for the current gait and the next cyclic gait are given, the initial state of the inverted pendulum of the cyclic gait is calculated, and the desired ZMP trajectory of the current gait is modified so that the current gait converges to the cyclic gait using the concept called “the divergent component of the motion” as a relaxed boundary condition. Using these techniques, real-time gait generation is achieved which requires a comparatively small amount of modification of the desired ZMP trajectory while guaranteeing the continuity of the gait generation.
