Maria Fox, Derek Long, Stephen Bradley, and James McKinna
Planning domain descriptions contain many structural features, not made explicit by the domain designer but nevertheless present in the representation, that can be exploited by a planner to reduce search. We have explored the effect of a range of logical features, extracted during pre-planning analysis, on the search performance of both forward- and backward-searching planning systems and have confirmed that considerable planning-time benefits can result from the use of efficient pre-planning analysis techniques. Domains involving time and other numeric quantities contain non-logical features which are harder to extract from a domain model. We have explored the combination of our existing pre-planning technology with standard model-checking approaches in order to extract temporal features of a planning domain that can improve planning performance. In this paper we explain how such domain features are extracted and how they can be exploited by the search strategy of a planner.