Supervisory control of hybrid systems based on model abstraction and guided search Original Research Article

This paper addresses the task of computing supervisory controllers by which hybrid systems with nonlinear continuous dynamics are driven into goal sets while safety specifications are met. For this class of systems, the (conservatively approximative) determination of reachable states is an important but also a computationally expensive step of the controller synthesis. This contribution proposes a technique aiming at reducing the reach set computation by using abstract models and guided search. For a discrete abstraction of the hybrid model, candidate paths are determined as possible controlled evolutions which fulfill the given specifications. A validation scheme comprising three different techniques is applied to determine whether the candidate path represents a feasible control strategy for the hybrid system. If the specification is violated, the abstract model is refined according to the validation result. The iterative application of the determination of candidate paths, the path validation and the model refinement steers the search for a control strategy such that often only a relatively small part of the reachable state space has to be explored. The synthesis procedure is illustrated for two examples.