Unified Foothold Selection and Motion Planning for Legged Systems in Real-Time

This work presents a novel architecture that unifies footstep planning, motion planning, and online feedback control for legged robots moving through complex environments. Our approach contrasts related prior works that treat planning and control as separate components in a hierarchical framework (first plan, then control). Though prior works have demonstrated success, existing state-of-the-art planning and control architectures for legged robots quickly become brittle in highly uncertain environments due to an inherent inability to dynamically and decisively react to unplanned events. To address this, this work presents a novel framework that uses modeling and analysis tools from the hybrid systems and nonlinear control communities to reformulate planning footsteps and dynamic trajectories as well as deriving closed-loop controllers as a single trajectory optimization problem. By combining these previously disparate steps we empirically show that we can remove much of complexity that underlies the hierarchical decision posed by conventional approaches, making it possible to dynamically and safely react to large external disturbances in sub-real-time. We present results that highlight the reactive and robust nature of the unified framework developed.