A Framework for Analyzing Resource / Quality Tradeo s inReal - Time AIDavid

Many researchers currently seek to combine the predictability of real-time systems with the intelligence of AI planning systems. We develop a framework for describing various approaches to this problem, by distinguishing strategic and tactical methods for meeting real-time performance guarantees. We introduce the any-dimension algorithm concept to clarify the fundamental connict between real-time and AI, showing that existing real-time systems techniques are incompatible with the goal of intelligent behavior. Using this framework, we examine several existing approaches to real-time AI, and motivate our hybrid approach| the Cooperative Intelligent Real-Time Control Architecture (CIRCA)| which combines aspects of strategic and tactical real-time methods to provide the exible yet predictable performance required for real-time AI. The opinions, ndings, and recommendations expressed in this publication are those of the authors, and do not necessarily reeect the views of the funding agencies. David Musliner is also aaliated with the UM Institute for Systems Research (NSF Grant NSFD CDR-88003012). ABSTRACT Many researchers currently seek to combine the predictability of real-time systems with the intelligence of AI planning systems. We develop a framework for describing various approaches to this problem, by distinguishing strategic and tactical methods for meeting real-time performance guarantees. We introduce the any-dimension algorithm concept to clarify the fundamental connict between real-time and AI, showing that existing real-time systems techniques are incompatible with the goal of intelligent behavior. Using this framework, we examine several existing approaches to real-time AI, and motivate our hybrid approach| the Cooperative Intelligent Real-Time Control Architecture (CIRCA)| which combines aspects of strategic and tactical real-time methods to provide the exible yet predictable performance required for real-time AI.