A study of synthesizing artificial intelligence (AI) planning domain models by using object constraints.

This paper concerns the area of automated acquisition of planning domain models from one or more examples of plans within the domain under study. It assumes that an adequate domain model for a domain can be composed of objects arranged in collections called object sorts. Recently, two systems have had success in using this underlying assumption: the Opmaker2 system (McCluskey et al. 2009), and the LOCM system (Cresswell, McCluskey, and West 2009). The former requires only one solution plan as input, as long as it contains at least one instance of each operator schema to be synthesized. It does require a partial domain model as well as the example plan, and the initial and goal states of the plan. In contrast LOCM requires no background information, but requires many instances of plans before it can synthesize domain models. Our aim is to build on these systems, and establish an experimental and theoretical basis for using object - centred assumptions to underlie the automated acquisition of planning domain models