Reaction-First Search

摘要

This paper presents Reaction-First Search (rfs), an incremental planning algorithm that produces plans for execution by a reactive system. The reactive system is independently competent in the sense that it is able to produce behavior in its environment without a plan. This reactive system has some prior probability of satisfying a given goal, and plans found by RFS serve to increase the probability of goal satisfaction above that prior. RFS is usefully incremental in that any partial plan it produces can be executed by the reactive system. While any given partial plan may or may not increase the reactive system's goal satisfaction probability, the expected performance of RFS is to monoton-ically increase the goal satisfaction probability as a function of time spent planning.