Synthesizing Efficient Controllers

In many situations, we are interested in controllers that implement a good trade-off between conflicting objectives, e.g., the speed of a car versus its fuel consumption, or the transmission rate of a wireless device versus its energy consumption. In both cases, we aim for a system that efficiently uses its resources. In this paper we show how to automatically construct efficient controllers. We provide a specification framework for controllers in probabilistic environments and show how to synthesize implementations from them. We achieve this by reduction to Markov Decision Processes with a novel objective function. We compute optimal strategies for them using three different solutions (linear programming, fractional linear programming, policy iteration). We implemented and compared the three algorithms and integrated the fastest algorithm into the model checker PRISM.