Artificial Intelligence in Reactive Scheduling

G. Haslea and S.F. Smithb aSection of Knowledge-based Systems, SINTEF Informatics, P.O. Box 124, 0314 Oslo, Norway b Center for Integrated Manufacturing Systems, The Robotics Institute, Carnegie Mellon University, Pittsburgh, PA 15213-3890, USA Opportunistic scheduling offers a uniform perspective on predictive and reactive scheduling as iterative problem solving processes. In the context of reactive scheduling, it constitutes a knowledge-directed alternative to more search intensive iterative approaches. By adopting a scheduling process that opportunistically focuses attention on the most critical subproblem and carefully selecting the focal point of the next problem solving effort, one can significantly constrain search while continuing to give attention to important scheduling objectives. Thus, one can maintain high-quality solutions in the face of changing constraints under stringent response time constraints. Control heuristics implemented in an architecture for opportunistic control determines the identification, analysis, and prioritization of subproblems, as well as the formulation of problem solving tasks. The nature of the control architecture determines the the span of control heuristics that may be accommodated. In addition to the repertoire and nature of methods for subproblem resolution, the nature of control heuristics plays a critical role in the performance of opportunistic scheduling systems. This paper describes a novel control architecture which represents a generalization of earlier architectures for opportunistic scheduling. It accommodates what we have denoted as focal point-opportunistic scheduling strategies. New control heuristics that draw upon the extended expressiveness of the novel control architecture are presented, as well as results from a comparative, empirical investigation of these heuristics based on reactive scenarios for a rich factory model. Keyword Codes: !.2.1; !.2.4; !.2.8