Robust Motion Planning Using a Maneuver Automaton with Built-in Uncertainties

In this paper, we extend a recently introduced motion planning framework for autonomous vehicles based on a maneuver automaton representation of the vehicle dynamics. We bring robustness into the guidance system by accounting for the uncertainties in the motion primitives used by the maneuver automaton. The uncertainties are taken into account in the offline computation of a guidance function, as well as in a real-time planning policy. We illustrate our approach using a high-fidelity simulation model of MIT's autonomous X-Cell miniature helicopter, and present an example that highlights the performance improvement over the original frame-work. We demonstrate that, when uncertainties are present, a nominal planning policy generates suboptimal trajectories in both open- and closed-loop guidance, and that trajectories obtained by applying the robust policy are less sensitive to perturbations in the motion primitives.