Active Learning of One-Clock Timed Automata using Constraint Solving
Active automata learning in the framework of Angluin's $L^*$ algorithm has been applied to learning many kinds of automata models. In applications to timed models such as timed automata, the main challenge is to determine guards on the clock value in transitions as well as which transitions reset the clock. In this paper, we introduce a new algorithm for active learning of deterministic one-clock timed automata and timed Mealy machines. The algorithm uses observation tables that do not commit to specific choices of reset, but instead rely on constraint solving to determine reset choices that satisfy readiness conditions. We evaluate our algorithm on randomly-generated examples as well as practical case studies, showing that it is applicable to larger models, and competitive with existing work for learning other forms of timed models.