Abstract
Mobile robot navigation is typically regarded as a geometric problem, inwhich the robot's objective is to perceive the geometry of the environment inorder to plan collision-free paths towards a desired goal. However, a purelygeometric view of the world can can be insufficient for many navigationproblems. For example, a robot navigating based on geometry may avoid a fieldof tall grass because it believes it is untraversable, and will therefore failto reach its desired goal. In this work, we investigate how to move beyondthese purely geometric-based approaches using a method that learns aboutphysical navigational affordances from experience. Our approach, which we callBADGR, is an end-to-end learning-based mobile robot navigation system that canbe trained with self-supervised off-policy data gathered in real-worldenvironments, without any simulation or human supervision. BADGR can navigatein real-world urban and off-road environments with geometrically distractingobstacles. It can also incorporate terrain preferences, generalize to novelenvironments, and continue to improve autonomously by gathering more data.Videos, code, and other supplemental material are available on our websitehttps://sites.google.com/view/badgr