Towards an Improved Domain Representation for Planning

Reference: P. Clark. Towards an improved domain representation for planning. Master's thesis, Edinburgh Univ., Edinburgh, UK, 1985.

Abstract: This dissertation is a study of the representation of domains used in AI planning systems. All planning systems must be provided with at least a minimal knowledge of the domain in which they are to operate, and the first part of the dissertation looks at the representation of this `essential' information. It proposes some extensions, including the use of four different goal types `achieve', `remove', `maintain' and `prevent', which enhance the descriptive power of the formalism, and the effectiveness of these extensions is demonstrated.

Most contemporary AI planning systems however do not just have this minimum knowledge, but also use additional domain-specific knowledge to guide and hence reduce the search. This makes planners more efficient but often at the expense of losing some of their generality. The second part of this dissertation addresses the problem of finding additional types of knowledge which are both useful and common to many domains, such that domain-specific knowledge can be used in the planning process without serious loss of generality. It surveys some of these additional types of knowledge already used by AI planning systems, and proposes some new types which are similarly both general and useful. The production rule formalism of expert systems is proposed and demonstrated as being a potentially powerful tool for the representation of such knowledge.

Postscript: Not available. Email me with your postal address if you would like a hard-copy.