Google’s Genie world model can now simulate real streets with Street View

Source: TechCrunch

Overview

Google DeepMind is now linking Street View to Project Genie, its general‑purpose world model that can generate diverse, interactive environments. Announced at the Google I/O developer conference, the integration lets users simulate real streets with adjustable conditions such as weather or time of year.

Integration with Project Genie

• Project Genie is DeepMind’s world model that creates interactive environments from text prompts or images.
• The new feature adds Street View imagery—over 280 billion images from 110 countries—to Genie, allowing simulations anchored to real locations.
• Initially available to Google AI Ultra subscribers in the United States, with a global rollout planned over the coming weeks.

“It’s really powerful for both the agent [and robotics] use case and for humans to play with, and that’s always been the thesis of Genie,” said Jack Parker‑Holder, a research scientist on DeepMind’s open‑endedness team.

Example Use Cases

• Robotics: Simulating rare lighting conditions (e.g., a sunny day in London) to prevent robots from being shocked by unexpected glare.
• Human Exploration: Viewing a New York City block under snowfall or other weather scenarios.
• Training Simulators: Waymo is already using Genie 3 to train self‑driving cars on rare events; adding Street View data could expand testing to more cities.

Technical Details

• Street View data has been collected for 20 years via camera‑equipped cars and “tracker backpacks,” covering virtually the entire globe.
• Genie combines this real‑world imagery with its generative capabilities, enabling viewpoint shifts from a car’s perspective to that of a human, robot, or other agents.
• Current simulations are video‑game quality rather than photorealistic and lack full physics awareness (e.g., characters can pass through objects).

Limitations

• Visual Fidelity: Results are recognizable but not photorealistic.
• Physics Awareness: Models do not yet understand cause and effect; objects can be intersected unrealistically.
• Accuracy Timeline: Parker‑Holder estimates the model is “six to 12 months behind video” in terms of quality and expects improvements soon.

Perspectives from Google

• Jonathan Herbert, director of Google Maps, notes that Genie cannot yet create a perfectly faithful street reconstruction, but its spatial continuity—remembering and simulating the environment behind a 360° turn—is a significant breakthrough.
• Diego Rivas, product manager at DeepMind, emphasizes that the Street View integration is still experimental, with ongoing work needed to improve accuracy.

Future Outlook

The goal is to make this capability widely accessible for educational experiences, gaming, robotics training, and more. As the model learns from passive observation, its understanding of physics and visual realism is expected to improve, bringing it closer to true photorealistic world simulation.