Topic: Physics-Based AI: Advancing Reliable Intelligence for Industrial Systems

AI has achieved impressive results across many domains, but ensuring reliability and trust remains a challenge, especially in industrial and safety-critical environments. This talk introduces physics-based AI, an approach that integrates machine learning with fundamental physical laws to enable more robust and dependable real-world systems. Unlike generative AI, which focuses on content creation, physics-based AI is designed to model and control physical systems and processes. By embedding principles from thermo fluid dynamics, electromagnetics, and mechanics, these approaches enforce physical consistency while leveraging data-driven insights. Drawing on recent research at MERL, the talk highlights advances that combine physics-based reasoning, advanced sensing, and machine learning to improve automation, efficiency, and reliability. Example applications include radar imaging, robotics, and energy-efficient data center operation through airflow sensing and optimization. Physics-based AI offers a path toward more trustworthy and energy-efficient solutions, and is poised to shape the next generation of innovation across manufacturing, robotics, and large-scale infrastructure.