Safety Verification of AI-enabled Cyber-Physical Systems

Title of the Talk: Safety Verification of AI-enabled Cyber-Physical Systems
Speakers: Lipsy Gupta
Host Faculty: Ramakrishna Upadrasta
Date: Dec 31, 2025
Time: 09:30 am.
Venue: CS-105

Abstract: AI-based components are increasingly integrated into autonomous and cyber-physical systems (CPS), replacing traditional modules in control and perception. These integrations have led to significant advances in safety-critical domains such as aerospace, automotive systems, and robotics. At the same time, the increased adaptability and complexity of AI-enabled components introduce new safety challenges, as failures in such systems can lead to catastrophic consequences. To address these challenges, formal verification is emerging as a powerful framework for providing guarantees on the correctness and safety of such systems.

In this talk, we focus on safety verification of AI-enabled CPS along two complementary directions. First, we address the continuously evolving nature of neural network controllers, which are frequently retrained or updated to improve performance. We introduce an approximate conformance framework that enables safety guarantees of a previously verified network to be transferred to its retrained variants without re-verifying the entire closed-loop system from scratch. Second, we consider safety verification of anytime perception-based CPS. Anytime sensors are being introduced to offer flexibility in latency and accuracy, unlike traditional sensors with fixed performance. Such variations in sensing performance can lead to mission-critical failures. We address this challenge by developing efficient reachable-set computation algorithms for closed-loop systems with anytime sensors and neural network controllers using the star set data structure and designing new algorithms for some star set operations.

Speaker’s Bio: Lipsy Gupta is a Research Assistant Professor in the Department of Computer Science at Kansas State University. She was previously a Postdoctoral Fellow at Kansas State University and at the University of Missouri–Columbia. Her current research interests are in the safety and reliability of AI-enabled cyber-physical systems, with a focus on neural network verification and formal methods for control and robotics. She has been recognized at Kansas State University with several awards including the Excellence in Research Award and the Outstanding Research Publication Award. Her research has been published in leading venues such as International Conference on Robotics and Automation (ICRA), International Conference on Embedded Software (EMSOFT), and Conference on Decision and Control (CDC). She is also actively involved in the research evaluation process, serving as a reviewer and subreviewer for leading conferences in formal methods and cyber-physical systems. She holds a Ph.D. in Mathematics from the Indian Institute of Technology Delhi, where her work centered on topology of metric spaces.

Meeting link: https://meet.google.com/ftp-fiqn-ief