Using the learned #attention to learn a policy over the pre-trained options in order to solve a downstream task, we observe that the agent with #affordances consistently outperforms the other agents. #NeurIPS2022 #Longhorizonsparserewards