The document discusses hierarchical reinforcement learning (HRL). It provides an overview of reinforcement learning and Markov decision processes before introducing HRL. HRL can improve performance by decomposing problems into subproblems using temporal and state abstraction. Several HRL approaches are described, including options which define subpolicies and termination conditions. The document outlines future work, such as automated discovery of subgoals and state abstractions, and developing agents that can continually learn across tasks.