Learning to Coordinate Control Policies of Hybrid Systems

Hybrid control systems consist of a discrete event system (DES) supervising the behaviour of a continuous state system (CSS) through the issuance of logical control directives. This paper derives sufficient conditions on the DES/CSS interface which guarantee the existence of a supervisor which transitions the plant through an arbitrary sequence of commanded events. It is further demonstrated that this interface can be learned using an inductive inference protocol which converges after a finite number of updates.