Department Seminars

Zoom link »

Institute for Assured Autonomy and Computer Science Seminar Series

“MAD Games: Multi-Agent Dynamic Games—Lessons from the Limits of Autonomous Racing”

Abstract: The critical challenge in deploying autonomous systems is achieving peak performance without compromising safety. Autonomous racing crystallizes this challenge, as it punishes timid policies and demands robust, adaptive strategies in multi-agent settings. Current approaches often fail by either oversimplifying the behavior of other agents or lacking mechanisms for real-time adaptation. This talk presents research that pushes the boundaries of perception, planning, and control. We will explore how to develop highly competitive agents through: 1) Adversarial training: Leveraging game theory and distributionally robust online adaptation to create agents that dynamically balance safety and assertiveness; 2) Adaptive safety: Using conformal prediction, control barrier function and imitation learning, we show how multiple imperfect experts train an AI to perform better than any single expert; 3) Learning invariant sets from failure and success: introduces an information-theoretic MPC framework using normalizing flows alongside a model-free method that directly carves out certified control-invariant sets from closed-loop failure data. All research is implemented on Mangharam’s RoboRacer.ai platform—1/10th the size, but 10x the fun. The key takeaway is a deeper understanding of how to build and validate safe autonomous systems for complex, interactive environments.

Speaker Biography: Rahul Mangharam is a professor of electrical and systems engineering and computer and information Science at the University of Pennsylvania. His research centers on trustworthy AI for safe autonomous systems. His work bridges formal methods, machine learning, and control theory to create provably safe systems for applications including autonomous vehicles, urban air mobility, and life-critical medical devices. For his contributions to life-critical systems, Mangharam received the Presidential Early Career Award for Scientists and Engineers from President Obama. His work has also been recognized with the NSF Early CAREER Award, the Intel Early Faculty Career Award, the Institute of Electrical and Electronics Engineers (IEEE) Benjamin Franklin Key Award, multiple best paper awards from the ACM and IEEE, and selection to the National Academy of Engineering’s Grainger Foundation Frontiers of Engineering program. Mangharam serves as the UPenn director for the $20 million Safety21 National University Transportation Center, a U.S. Department of Transportation center for safe and efficient mobility. He also directs the Autoware Center of Excellence, an open-source autonomous driving consortium of over 33 academic partners, and is the founder of the RoboRacer (formerly F1TENTH) Autonomous Racing Community, now active in over 90 universities worldwide.

Computer Science and Information Security Institute Seminar Series

November 4, 2026

Abstract: Cyber-physical systems (CPS) tightly couple computing and network components with physical processes via sensors and actuators. On the one hand, integration of AI enables autonomous CPS that perceive, understand, and perform complex tasks in the physical world. On the other hand, safety is critical for real-world CPS applications such as autonomous vehicles, drones, and various robotic systems. In this talk, Fanxin Kong will introduce his recent works on assured and intelligent CPS as follows: i) Foundation model-enabled CPS—this thread of works studies how to enhance safety of foundation models (e.g., large language models and vision-language action models) when applied to task/motion planning; ii) Secure and safe reinforcement learning (RL)—this thread of works addresses the synthesis of safe control policies using RL and explores safety-violation vulnerabilities of safe RL; and iii) Real-time attack-resilience for CPS—this thread of works discusses how to detect, diagnose, and recover from sensor faults/attacks in real-time and in a safe manner. System demonstrations with implementation of these works on multiple autonomous CPS simulators/testbeds will be also presented.

Speaker Biography: Fanxin Kong is an assistant professor in the Department of Computer Science and Engineering at the University of Notre Dame. His research centers around assured and intelligent cyber-physical systems with a focus on physical AI, safety and security, and real-time embedded systems, as well as their applications to various robotic systems. Kong has published over 85 research papers at top venues such as the Institute of Electrical and Electronics Engineers (IEEE) International Conference on Robotics and Automation, the ACM/IEEE International Conference on Cyber-Physical Systems, the IEEE Symposium on Real-Time Systems, the IEEE Real-Time and Embedded Technology and Applications Symposium, the IEEE/ACM International Conference on Embedded Software, and the Design Automation Conference; in various IEEE/ACM transactions; and books and book chapters. His research has been supported by the NSF, the Air Force Research Laboratory (AFRL), the Air Force Office of Scientific Research, and DARPA. He has received multiple awards such as the 2025 NSF CAREER Award, the 2025 ACM Special Interest Group on Embodied Systems (SIGBED) Early Career Researcher Award, and the AFRL Summer Faculty Extension Award in 2022 and 2024. Together with his students, Kong has won multiple competitions such as the 2024 Embedded System Software Competition at Embodied Systems Week and the 2024 ACM SIGBED Student Research Competition (SRC), as well as taking home the Best Scientific Research Award at the ACM SIGBED SRC in 2022.