Supervisor Localization of Discrete-Event Systems under Partial Observation
For researchers in discrete-event systems, this provides a method to distribute control under partial observation, but it is an incremental extension of prior work.
This paper extends supervisor localization for discrete-event systems to handle partial observation, using relative observability to compute a monolithic supervisor and then decomposing it into local controllers that only change state on observable events. The approach is combined with heterarchical synthesis for scalability.
Recently we developed supervisor localization, a top-down approach to distributed control of discrete-event systems. Its essence is the allocation of monolithic (global) control action among the local control strategies of individual agents. In this paper, we extend supervisor localization by considering partial observation; namely not all events are observable. Specifically, we employ the recently proposed concept of relative observability to compute a partial-observation monolithic supervisor, and then design a suitable localization procedure to decompose the supervisor into a set of local controllers. In the resulting local controllers, only observable events can cause state change. Further, to deal with large-scale systems, we combine the partial-observation supervisor localization with an efficient architectural synthesis approach: first compute a heterarchical array of partial-observation decentralized supervisors and coordinators, and then localize each of these supervisors/coordinators into local controllers.