Synthesis of Supervisory Controllers for Hybrid Systems Using Abstraction Refinement

For the computationally challenging task of synthesizing supervisory controllers for hybrid systems, this paper suggests to use guided abstraction refinement. Starting from a discrete abstraction of the hybrid model, so-called 'candidate paths' are determined as possible controlled evolutions which satisfy given specifications for safety and goal attainment. A specialized validation procedure is used to check if the corresponding control strategies satisfy the specifications also for the original hybrid system. If not, the abstract model is refined according to the validation result. Since different validation methods (with different computational costs) are used, reachable hybrid sets have to be computed only for relatively few combinations of control inputs and locations of the hybrid automaton.