Towards inducing hierarchical task network domain models for AI planning from examples

Domain modelling for AI Planning aims to form a database of facts about the ‘world’ being modelled.
This can be a complex process especially if there is a large number of objects or actions or both to be
modelled. This task can be facilitated by tools which induce operators or methods from examples.
Further, large and complex domains are more easily constructed if domain languages are used which
allow for hierarchical decomposition of domain components. Examples of such a decomposition are
class hierarchies and method hierarchies. This paper describes ongoing work which aims to
produce algorithms which learn effective hierarchical decompositions from examples.