{"title":"Using control synthesis for falsification and corner case generation","authors":"N. Ozay","doi":"10.1145/3459086.3459632","DOIUrl":null,"url":null,"abstract":"This talk will describe algorithms that search for \"dynamical adversarial examples\" or \"corner cases\" for feedback control systems. This problem is related to the falsification problem, where the goal is to find initial conditions, disturbance profiles, and environment behaviors that force the system to violate its specifications. As opposed to the commonly adopted falsification approaches that treat the system under test as a black-box, we propose a synthesis-guided approach, which leverages the knowledge of a plant model if it exists and treats only the controller and perception mechanism as black-box. Our algorithm uses the plant model and backward reachable set computations to guide the search for falsifying trajectories. We will demonstrate the approach with examples from autonomous systems, including those using perception-based neural network controllers.","PeriodicalId":127610,"journal":{"name":"Proceedings of the 1st International Workshop on Verification of Autonomous & Robotic Systems","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2021-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the 1st International Workshop on Verification of Autonomous & Robotic Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3459086.3459632","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
This talk will describe algorithms that search for "dynamical adversarial examples" or "corner cases" for feedback control systems. This problem is related to the falsification problem, where the goal is to find initial conditions, disturbance profiles, and environment behaviors that force the system to violate its specifications. As opposed to the commonly adopted falsification approaches that treat the system under test as a black-box, we propose a synthesis-guided approach, which leverages the knowledge of a plant model if it exists and treats only the controller and perception mechanism as black-box. Our algorithm uses the plant model and backward reachable set computations to guide the search for falsifying trajectories. We will demonstrate the approach with examples from autonomous systems, including those using perception-based neural network controllers.