Finding Equivalences Among Abstract Actions

Separating what is relevant from what is not can turn an intractable learning problem into a solvable one by reducing the complexity of the problem. Abstraction is a good general tool for sorting out relevant and irrelevant information. There are two common types of abstraction: state abstraction, and hierarchical actions. This paper examines the ways in which state abstraction can be used in the context of a given action hierarchy.