Manipulate-to-Navigate: Reinforcement Learning with Visual Affordances and Manipulability Priors