Chapter 10 - Control Rules in Planning

Abstract

The efficiency of a planning procedure can often be improved dramatically if a good way can be found to do this—for example, in some cases, it may enable a procedure to solve problems in polynomial time that might otherwise take exponential time. Deciding whether a node can be pruned often involves a collection of highly domain-specific tests to detect situations in which one can be confident that the solutions that lie below a node are less desirable than the solutions that lie below nodes elsewhere in the search space. To write pruning rules, a language is needed for expressing relationships among properties of the current state and properties of subsequent states. This chapter describes such a language and a planning algorithm based on it. The presentation is based loosely on ideas developed by Bacchus and Kabanza. The chapter describes a language for writing control rules, a procedure for evaluating control rules written in that language, and a planning procedure based on the inference procedure. The chapter also describes ways to extend the approach. The final section discusses how to handle certain kinds of extended goals.