Model-based RL agents today ... to learn both the nodes and the edges on the graph together with a goal-conditioned policy, and how to better leverage temporal abstraction in online planning.