Beyond single shooting: Iterative approaches to falsification
Jyotirmoy V. Deshmukh, Georgios Fainekos, James P. Kapinski, Sriram Sankaranarayanan, Aditya Zutshi, Xiaoqing Jin · 2015
Summary form only given. Simulation-based falsification techniques using robustness-guided stochastic search have been successful in finding property violations in systems that are otherwise not amenable to formal analysis. A central problem, however, lies in the phenomenon of simulation-explosion: wherein the number of simulations required to find violations increases with the size of the system, along with the time taken for each simulation. Another problem arises due to corner case phenomenon for hybrid systems wherein violations are found due to particular initial conditions, input and switching sequences, that are very hard to find using stochastic search. In this talk, we demonstrate techniques that leverage fewer and/or shorter simulation traces by modifying the system and the properties to iteratively converge towards a falsification of the original property for the original system. We present the idea of trajectory splicing, that explores and incrementally refines multiple, short trajectory segments to yield a falsification of the original system. Next, we present the notion of source/target enlargement that makes corner case bugs easier to find by expanding the set of initial conditions, the target set and the systems switching conditions to make the process of finding falsifications easier. We will briefly address requirements in simulation technologies for large systems that can make the algorithms presented work faster and more effectively for model-based designs.