Abstract: Despite their increasing sophistication, modern robotic systems still struggle to operate safely and reliably around people in uncertain open-world situations, a key bottleneck to adoption that is perhaps best exemplified by the growing public distrust of early autonomous driving technology. At the same time, the booming excitement around generative AI models has been accompanied by new concerns about the individual and societal harms that may result from poorly understood human–AI interaction over time, with current guardrails being notoriously easy to bypass. Safety assurances are proving elusive in both domains, but what if each contained the key to the other?

This talk will discuss a new window of opportunity to bridge the rapidly growing capabilities of generative AI and the dynamical safety frameworks from robotics and control theory, laying a common foundation for safe human-centered autonomy in the coming decades. We will first review recent advances in *safety filters*, real-time algorithms that monitor an autonomous system’s operation and intercept actions that could lead to future failures, and show how adversarial reinforcement learning can be used to systematically synthesize filters with clear-cut guarantees in previously intractable robotics problems, from traversing abrupt terrain to sharing the road with people. We will then see how AI systems equipped with large web-trained models can use introspective self-prompting to refine their situational uncertainty and identify potential safety hazards even when they have never encountered them before. With these insights, we will outline the key components and technologies needed for general human–AI safety filters that can monitor AI systems across a wide range of domains from driving to natural language, anticipate different forms of danger, and proactively steer interaction towards safe outcomes.

Bio: Jaime Fernández Fisac is an Assistant Professor of Electrical and Computer Engineering at Princeton University, where he directs the Safe Robotics Laboratory and co-directs the Princeton AI4ALL summer program. His research combines control theory and artificial intelligence with psychology and game theory with the goal of enabling robots to operate safely around people in a way that is well understood, and thereby trusted, by their users and the public at large. Prior to joining Princeton, he was a Research Scientist at Waymo (Google’s self-driving sister company), where he was involved in launching new technical thrusts on safe interaction. He is also the co-founder of Vault Robotics, a startup developing safe and agile van-to-door robots for the last 50 feet of package delivery. Prof. Fisac received an Engineering Diploma from the Universidad Politécnica de Madrid, Spain in 2012, a Master’s in Aeronautics from Cranfield University, U.K. in 2013, and a Ph.D. in Electrical Engineering and Computer Sciences from the University of California, Berkeley in 2019. His work has been featured by the Wall Street Journal, NBC, WIRED magazine, and the Robohub podcast, funded by DARPA and the NSF, and recognized with the La Caixa Foundation Fellowship, the UC Berkeley Leon O. Chua Award, the Google Research Scholar Award, and the Sony Focused Research Award.