{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/verifying-controllers-against-adversarial","title":"Verifying Controllers Against Adversarial Examples with Bayesian Optimization","arxiv_id":"1802.08678","date":"2018-02-23","proceeding":null,"authors":["Shromona Ghosh","Felix Berkenkamp","Gireeja Ranade","Shaz Qadeer","Ashish Kapoor"],"abstract":"Recent successes in reinforcement learning have lead to the development of\ncomplex controllers for real-world robots. As these robots are deployed in\nsafety-critical applications and interact with humans, it becomes critical to\nensure safety in order to avoid causing harm. A first step in this direction is\nto test the controllers in simulation. To be able to do this, we need to\ncapture what we mean by safety and then efficiently search the space of all\nbehaviors to see if they are safe. In this paper, we present an active-testing\nframework based on Bayesian Optimization. We specify safety constraints using\nlogic and exploit structure in the problem in order to test the system for\nadversarial counter examples that violate the safety specifications. These\nspecifications are defined as complex boolean combinations of smooth functions\non the trajectories and, unlike reward functions in reinforcement learning, are\nexpressive and impose hard constraints on the system. In our framework, we\nexploit regularity assumptions on individual functions in form of a Gaussian\nProcess (GP) prior. We combine these into a coherent optimization framework\nusing problem structure. The resulting algorithm is able to provably verify\ncomplex safety specifications or alternatively find counter examples.\nExperimental results show that the proposed method is able to find adversarial\nexamples quickly.","url_abs":"http://arxiv.org/abs/1802.08678v2","url_pdf":"http://arxiv.org/pdf/1802.08678v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"verifying-controllers-against-adversarial","repo_url":"https://github.com/shromonag/adversarial_testing","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"bayesian-optimization","task_name":"Bayesian Optimization"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}