Bipedal running with nearly-passive flight phases

This paper presents a novel approach which generates running gaits for a bipedal robot model. Inspired by human running, we assume that the flight phase is nearly-passive. Thus, the initial and final joint velocities follow from a static optimization procedure, provided that the boundary joint angles have been picked in advance. Dynamic optimization then produces the two running phases, with a simple collision model predicting the discrete joint velocity changes. Ground reaction forces, the Zero-Moment Point, and foot clearance of the swing leg serve as nonlinear constraints. A biped robot running in simulation demonstrates the effectiveness of the proposed approach.