CityRAG: Stepping Into a City via Spatially-Grounded Video Generation
Researchers have developed CityRAG, a novel video generation model capable of creating navigable, 3D-consistent environments grounded in real-world geography. This model can generate long video sequences that maintain consistent weather and lighting conditions, even with temporally unaligned training data. CityRAG's ability to disentangle scene structure from transient attributes allows for applications in autonomous driving and robotics simulation. AI
IMPACT Enables creation of realistic, navigable 3D environments for training autonomous systems.