Using Reactive Rules to Guide a Forward-Chaining Planner
Murray Shanahan
Abstract
This paper presents a planning technique in which a flawed set of reactive
rules is used to guide a stochastic forward-chaining search. A planner based
on this technique is shown to perform well on Blocks World problems. But
the attraction of the technique is not only its high performance as a straight
planner, but also its anytime capability. Using a more dynamic domain, the
performance of a resource-bounded version of the planner is shown to degrade
gracefully as computational resources are reduced.