Spatial AI and World Models Toward General Physical Intelligence

This panel discussion features robotics industry leaders exploring the transition of artificial intelligence from digital domains to physical world automation. Speakers Kei Kase (Nvidia Japan), Udaya Rongala (IntuiCell), Ali Agha (Field AI), and Ivan Poupyrev(Archetype AI) discuss how spatial AI and world models serve as the foundational architecture for general physical intelligence. The presentation details the critical challenges of physical AI deployment, including limited robotic datasets, distribution drift, sensor fusion, and mitigating transformer hallucination in unstructured environments. Finally, the panel evaluates the compute restrictions of edge devices and offers varying timelines on when autonomous humanoids will achieve general physical versatility.