Google DeepMind Adds Street View Data to Genie World Model

Google DeepMind is integrating Street View data with Project Genie to generate interactive simulations of real-world locations. The feature launched during the Google I/O developer conference and is now available to some Google AI Ultra subscribers in the United States.

The system lets users adjust conditions such as weather or time of year within a simulated environment. Jack Parker-Holder, a research scientist on DeepMind’s open-endedness team, said the tool could help robots encounter rare conditions like sunlight on Victorian housing in London.

Google has collected more than 280 billion Street View images across 110 countries over 20 years. Parker-Holder said combining this imagery with Genie’s simulation abilities creates a powerful source of real-world information. Genie 3, released for research preview last August, is already used in one of Waymo’s simulators to train self-driving cars on rare events.

Street View data could expand those simulations beyond the vehicle’s perspective to include other viewpoints such as pedestrians or robots.

The models are not yet physics-aware, and results remain at video-game quality rather than photorealistic. Parker-Holder said accuracy and quality are expected to improve within six to 12 months. Access is rolling out to global Google AI Ultra subscribers over the next few weeks. Diego Rivas, a product manager at DeepMind, said the feature remains an experiment with further improvements planned.