On-the-fly model abstraction for controller synthesis

While abstractions of dynamical systems render the rigorous and algorithmic synthesis of controllers feasible in principle, the computation of discrete abstractions is known to be feasible only for systems of relatively low dimension. As a step towards making the computations more tractable, this paper proposes to reduce the computational load and the required memory by combining the computation of the abstraction and the solution of an auxiliary discrete control problem. Although the worst-case complexity of the proposed algorithm is worse than for other known procedures, the algorithm shows promising performance for a challenging example.