Reachability-Aware Laplacian Representation in Reinforcement Learning