Abstract: How can intelligent agents truly navigate the real world—not just move within it? Embodied navigation seeks to answer this question by integrating perception, reasoning, and action into a unified framework for goal-directed behavior. This talk explores how we can make such navigation robust, generalizable, and adaptive, even in unseen, dynamic environments. We begin with the evolution from Object Goal Navigation to Zero-Shot Embodied Navigation, where robots learn to infer object locations they have never encountered before through vision-language foundation models (VLFMs) that blend spatial understanding with commonsense reasoning. I will then highlight our recent advances—multi-scale geometric-affordance guidance for grounding semantics in geometry, and diversified expert frontier exploration for more reliable decision-making under uncertainty. In the second part, we move beyond categories to Instance Image-Goal Navigation, enabling agents to locate a specific object instance (“my red mug”) via a novel 3D Gaussian Splatting–based hierarchical scoring framework that unifies appearance, geometry, and semantics for real-time localization. Finally, I will introduce Navigation Dreamer, a generative approach where the agent “imagines” unobserved space through MapBERT, creating semantic world models for proactive, long-horizon planning. Together, these efforts chart a path toward truly embodied intelligence—robots that see, think, and move with purpose in the open world.
Bio: Yi Fang is a professor, engineer, and entrepreneur specializing in Embodied AI and Robotics. He directs the NYU Embodied AI and Robotics (AIR) Lab, where he leads research at the intersection of perception, reasoning, and action — advancing AI-driven systems with applications across engineering, healthcare, and beyond. Beginning in 2026, he will serve as Co-Director (AI Track) of both the Center for AI and Robotics (CAIR) and the Center for Interdisciplinary and Data Science at NYU Abu Dhabi. Yi Fang earned his B.S. and M.S. degrees from Xi’an Jiaotong University and his Ph.D. from Purdue University. His contributions to AI and robotics have been recognized by the U.S. National Science Foundation (NSF) and Abu Dhabi’s Department of Education and Knowledge (ADEK). In addition to his academic leadership, he is the founder of ChatSign Technology, a company developing generative AI–powered assistive devices that enhance communication for Deaf and Hard-of-Hearing individuals, and EZ247 AI, an agentic AI company that connects customers and service providers with greater efficiency. Yi Fang actively contributes to the broader AI community, serving as an Area Chair or Senior Area Chair for top conferences including NeurIPS, CVPR, ICCV, ECCV, IROS, and IJCAI. His work bridges cutting-edge research and real-world innovation, reflecting a deep commitment to advancing AI for the benefit of society.