Reference: Hayes-Roth, B.; Brownston, L.; & Collinot, A. Strategic Control of Reactive Behavior in Intelligent Agents. Knowledge Systems Laboratory, February, 1993.
Abstract: An intelligent agent must often choose among several competing actions. To act effectively in real time, it must both generate strategic plans and respond to external events. Construction plans completely in advance lacks flexibility, while purely reactive systems can lack generality. We present an algorithm which dynamically constructs a control plan and yet uses this plan in a reactive fashion when response time is critical. This Satisficing algorithm uses the control plan to process both events and known actions in a most-promising-first order, terminating its search whenever it has either encountered a potential action that exceeds a declared threshold, reached a temporal deadline, or exhausted all known actions, whichever comes first. Experimental results are presented to demonstrate that the satisficing algorithm has the expected and desired characteristics.