On Comparing Planning Domain Models

Whereas research into the characteristics and properties of AI planning algorithms is overwhelming, a similar science of domain model comparison and analysis is underdeveloped. It has long been acknowledged that there can be a range of di�erent encodings for the same domain, whatever coding language used, but the question of which is the \best" encoding is an open one, and partly dependent on the requirements of the planning application itself. There is a growing need to measure and compare domain models, however, in particular to evaluate learning methods. In this paper we motivate the research by considering the challenges in evaluating domain model learning algorithms. We describe an ongoing doctoral research project which is exploring model classifications for comparing domain models.