Autonomous self-righting using recursive Bayesian estimation to determine unknown ground angles

Abstract

As robots are deployed to dynamic, uncertain environments, their ability to discern key aspects of their environment and recover from errors becomes paramount. In particular, tip-over events can potentially end or substantially disrupt mission performance and jeopardize asset recovery. To facilitate recovery from tip-over events (i.e. self-righting), the robot should be able to discern the ground angle on which it lies even when it is not in its preferred upright orientation. In this paper, we present a methodology for determining unknown ground angles using recursive Bayesian estimation. First, we briefly review our previous framework for autonomous self-righting, which we use to generate conformation space maps correlating stable robot configurations and orientations on various ground angles. Using these maps, we compare sensor orientation to predicted orientation for the robot configuration on all mapped ground angles. We then compute the best fit ground angle and assign it a confidence level based on filters such as predicted stability margin and measured rate of orientation change. We compare ground angle prediction error as a function of time using a variety of methods, and show a sensitivity analysis comparing accuracy as a function of the discretization of the ground angle dimension of the conformation space map. Finally, we demonstrate a physical robot's ability to self-right on unknown ground using this methodology.