ABSTRACT: Artificial Intelligence is fundamentally reshaping modern autonomous systems by integrating intelligent planning and reasoning into their operational capabilities. However, while this AI integration enhances the generalizable operation of such systems, it also introduces two novel challenges: (i) safety and (ii) security. While safety is a broad concern, we focus specifically on physical safety amid uncertainties arising from operating in unseen environments and sensor degradation. This talk will explore effective strategies to mitigate these uncertainties. Our first study investigates sensor fusion techniques to reduce ambiguity caused by sensor degradation. The second study examines multi-agent collaborative autonomy, which enhances safe traversability in unknown environments. Finally, the third study addresses human-robot teaming in construction sites, where rapidly changing and hazardous conditions introduce significant operational uncertainty. On the security front, the full stack of AI-integrated autonomy, spanning from perception through to planning, is highly vulnerable to adversarial attacks. These threats manifest as input-level adversarial perturbations, which range from subtle digital noise to physically-realizable artifacts such as adversarial patches and malicious electronic signals designed to deceive the system’s perception. We provide an overview of the spectrum of adversarial threats targeting AI-integrated autonomous systems, from the initial perception stage to the final planning and control stage. Additionally, we detail how these attacks successfully propagate through the entire system’s pipeline, adversely affecting downstream planning and control outputs, and ultimately leading to the hijacking of the system’s autonomy.

BIOSKETCH: Mohammad Al Faruque is currently a Conexant-Broadcom Endowed Chair Professor at the University of California, Irvine (UCI). He serves as the Director of the Center for Resilient Autonomous Systems. He is currently the ACM Distinguished Speaker and was the IEEE CEDA Distinguished Lecturer. Dr. Al Faruque’s research interests include AI-integrated autonomous systems, physical AI, cyber-physical security, and edge AI. He is the recipient of numerous awards, including the UCI School of Engineering Mid-Career Faculty Award for Research (2019), the IEEE Technical Committee on Cyber-Physical Systems Early-Career Award (2018), the IEEE CEDA Ernest S. Kuh Early Career Award (2016), the UCI Academic Senate Distinguished Early-Career Faculty Award for Research (2017), 4 best paper awards, including ACSAC, DAC, DATE, and ICCAD, the VehicleSec Best Demo Award (2025), and the NDSS Distinguished Poster Award (2016). He is also an IEEE Fellow (2026), AAIS Fellow (2026), and a Hellman Fellow (2015). In addition to publishing more than 200 IEEE/ACM papers in premier journals and conferences, Dr. Al Faruque holds 17 U.S. patents and has authored two books in the area of embedded and cyber-physical systems.