An Investigation into Distributed Constraint-directed Factory Scheduling

Abstract

An approach is presented to focus searches in a distributed system in individual agent search spaces so as to optimize decisions in the global space. The chosen domain is distributed-factory scheduling. Distributed decision-making in factory environments is important because factories are inherently distributed and because they need to effectively respond to change. The approach relies on a set of texture measures that quantify several characteristics of the space being searched. These texture measures play four important roles in a distributed search: (1) they focus the attention of an agent to globally critical decision points in its local search space, (2) they provide guidance in making a particular decision at a decision point, (3) they are good predictive measures of the impact of local decisions on system goals, and (4) they are used to model beliefs and intentions of other agents. The development of the texture measures is the result of extensive experimentation in a single-agent setting. The implementation of a distributed testbed is described. Experiments involving multiple agents are currently being performed.