This paper presents a method for synthesis of control alphabet policies, given continuum descriptions of physical systems and tasks. First, we describe a model predictive control scheme, called switched sequential action control (sSAC), that generates global state-feedback control policies with low computational cost, given a control alphabet. During synthesis, sSAC alphabet policies are directly encoded into finite state machines using a cell subdivision approach. As opposed to existing automata synthesis methods, controller synthesis is based entirely on the original nonlinear system dynamics and thus does not rely on but rather results in a lower-complexity symbolic representation. The method is validated for the cart-pendulum inversion problem and the double-tank system. The approach presents an opportunity for real-time task-oriented control of complex robotic platforms using exclusively sensor data with no online computation involved.