Synthesis of Safe, Optimal and Compact Strategies for Stochastic Hybrid Games (Invited Paper)

摘要

UPPAAL-Stratego is a recent branch of the verification tool UPPAAL allowing for synthesis of safe and optimal strategies for stochastic timed (hybrid) games. We describe newly developed learning methods, allowing for synthesis of significantly better strategies and with much improved convergence behaviour. Also, we describe novel use of decision trees for learning orders-of-magnitude more compact strategy representation. In both cases, the seek for optimality does not compromise safety.