Strategic refinements in abstraction based supervisory control of hybrid systems

Abstract

A common approach to hybrid control problems is to alternate refinement of a plant abstraction with trial controller synthesis performed on the current abstraction. These steps are repeated until a solution to the control problem can be found, or computational resources are exhausted. In this paper we use a temporal decomposition of the control problem in order to gain relevant diagnostic information for those steps when synthesis fails. We use this information to focus the abstraction refinement on those features of the plant that are most crucial for the synthesis task at hand. This work is an advance over earlier abstraction based synthesis procedures which use an unfocused and global refinement on the plant model.