Abstract

Omnidirectional vision, using 360-degree vision to understand theenvironment, has become increasingly critical across domains like robotics,industrial inspection, and environmental monitoring. Compared to traditionalpinhole vision, omnidirectional vision provides holistic environmentalawareness, significantly enhancing the completeness of scene perception and thereliability of decision-making. However, foundational research in this area hashistorically lagged behind traditional pinhole vision. This talk presents anemerging trend in the embodied AI era: the rapid development of omnidirectionalvision, driven by growing industrial demand and academic interest. We highlightrecent breakthroughs in omnidirectional generation, omnidirectional perception,omnidirectional understanding, and related datasets. Drawing on insights fromboth academia and industry, we propose an ideal panoramic system architecturein the embodied AI era, PANORAMA, which consists of four key subsystems.Moreover, we offer in-depth opinions related to emerging trends andcross-community impacts at the intersection of panoramic vision and embodiedAI, along with the future roadmap and open challenges. This overviewsynthesizes state-of-the-art advancements and outlines challenges andopportunities for future research in building robust, general-purposeomnidirectional AI systems in the embodied AI era.