What is AMDPs?

AMDPs, or Abstract Markov Decision Processes, are a framework for hierarchical reinforcement learning. They provide a way to decompose complex tasks into simpler, reusable sub-tasks. An AMDP represents a hierarchy of decision-making processes where each level consists of an MDP. High-level MDPs can call lower-level AMDPs as subroutines. This allows agents to learn abstract strategies and generalize across different environments by reusing learned skills.