Optimal control with regular objectives using an abstraction-refinement approach