Abstract control rules and their use in domain independent planning

Control rule based planners that take advantage of hand coded domain specific control knowledge currently out-perform fully automatic planners. Generic types have been presented as abstractions of the behaviour of types identified in planning domains and we extend this abstraction to sets of interacting generic types, or generic clusters. A state based modal temporal logic is presented for expressing properties of sequences of states, the terms of which are features of generic clusters. The logic enables control strategies to be written in terms of generic clusters. Automatic generic cluster identification provides a way of automatically specialising expressions with the details of any domain level instances, yielding domain specific control rules. We show how control rules employed by current control rule based systems can be expressed in the framework presented and also demonstrate generic control rules being used to improve plan quality in a state of the art, fully automatic planning system.