Reducing the Semantic Gap between Application Domain and AI Planning Technology: a compilation based PSM

We present a compilation-based approach that combines a formalism designed to match the structure of human expertise in the construction industry with an established planning technique. The domain's modelling requirements are derived from a study carried out with domain experts. This study shows that expertise in the industry is centred on the components of a building organised into a subcomponent structure. Encoding the results of this study into a HTN formalism demonstrates that such formalism fragment expert knowledge. This fragmentation leads to a large semantic distance between expert and formalism, making the task of encoding and maintaining a planner knowledge base a complex one. Our solution is to provide a formalism designed around the modelling requirements of the constriction industry and to then compile HTN schemas from that representation. We argue that this union reduces the semantic gap between knowledge engineer and formalism, thus lowering the complexity of the knowledge encoding and maintenance tasks, whilst still exploiting powerful AI planning techniques. We conclude by encouraging further work of this type with the aim of providing a library of domain oriented formalisms form which the knowledge engineer may choose an appropriate representation for a domain.