On Time with Minimal Expected Cost!

(Priced) timed games are two-player quantitative games involving an environment assumed to be completely antogonistic. Classical analysis consists in the synthesis of strategies ensuring safety, time-bounded or cost-bounded reachability objectives. Assuming a randomized environment, the (priced) timed game essentially defines an infinite-state Markov (reward) decision proces. In this setting the objective is classically to find a strategy that will minimize the expected reachability cost, but with no guarantees on worst-case behaviour. In this paper, we provide efficient methods for computing reachability strategies that will both ensure worst case time-bounds as well as provide (near-) minimal expected cost. Our method extends the synthesis algorithms of the synthesis tool Uppaal-Tiga with suitable adapted reinforcement learning techniques, that exhibits several orders of magnitude improvements w.r.t. previously known automated methods.