Predicted Step Viability: a stability criterion for biped gait

Abstract

The formulation of a stability criterion is crucial to perform safe, versatile and efficient biped gaits for robots. We here present a stability criterion for non-cyclic gait synthesis, the Predicted Step Viability, inspired by human gait and N-Step Capturability. The Predicted Step Viability defines the constraints of the current step such that future steps will be able to guarantee convergence to a stable point in finite time. In this way, it is based on the prediction of future viable steps to ensure stability. The criterion was implemented using multiphase trajectory optimization on two biped models, the Compass Gait and the five-link model RABBIT. The Compass Gait was simulated with different model parameters and gait patterns including a random non-periodic one. The five-link model was tested with linear and random reference gait patterns. The Predicted Step Viability criterion successfully generated stable non-periodic gaits under a variety of conditions. Moreover, it is possible to prescribe any gait pattern completely uncoupled from the stability criterion. If it were impossible to follow the prescribed pattern without falling, the controller would give it up, maintaining stability.