Google Genie Now Simulates Real Streets, India's Robotics on Notice
DeepMind has plugged Street View into Genie 3, letting users walk through real cities as interactive simulations. For Indian robotics startups, the implications run deep.
The News
Google DeepMind on Tuesday announced that its Genie 3 world model can now stitch together Street View imagery to produce interactive simulations of real-world streets. The feature was unveiled at Google I/O 2026 and is rolling out first to Google AI Ultra subscribers in the United States, with a wider release scheduled for the coming weeks.
The integration draws on a Street View archive that Google has built over two decades. By the company's count, more than 280 billion images have been captured across 110 countries and seven continents. Genie 3, which began life as a research preview in August 2025 and opened to AI Ultra customers in January 2026, can use that imagery to recreate roads, buildings, and neighbourhoods that a user can walk through, alter, and stress-test under different weather scenarios.
Jack Parker-Holder, a research scientist on DeepMind's open-endedness team, conceded the simulations were still rough. The output sits at what he described as "video game quality" rather than photorealism, and the model lacks physics awareness, meaning objects do not collide or react in lifelike ways. He estimated the model was "six to 12 months behind video in terms of accuracy". The feature has been shaped by Diego Rivas, a product manager at DeepMind.
Why It Matters
World models that can ingest real geography are a stepping stone toward cheaper robotics and autonomous-vehicle training. Until now, a Waymo or a Cruise has had to lean on costly purpose-built simulation engines and millions of miles of recorded driving data. A general-purpose simulator that draws on existing Street View images shifts a large chunk of that cost onto Google's pre-existing infrastructure.
This is the same playbook Google ran with Maps fifteen years ago, when it gave away navigation and starved standalone GPS makers. The historical parallel that matters here is OpenAI's Sora release in February 2024, which compressed high-fidelity video generation into a commodity within months. World models look set to follow the same curve, and Genie 3 is now the closest public demonstration of that trajectory.
Indian Angle
For Indian companies, the immediate question is data parity. Google withdrew Street View from India in 2011 over security objections and only returned in 2022 under a partnership with Tech Mahindra and Genesys International, which has since mapped 10 cities plus a handful of tourist circuits. The Indian street footprint inside Genie 3 will therefore be a fraction of what users in Tokyo or New York can simulate, and that gap will harden any local advantage held by domestic mapping outfits such as MapmyIndia and Genesys.
The Indian robotics and electric-mobility cohort, which includes Ather Energy, Ola Electric's autonomy unit, Niqo Robotics, and GreyOrange, has been building simulation stacks in-house. A general-purpose simulator bundled into AI Ultra tilts the build-versus-buy calculation. Expect founders to begin running side-by-side benchmarks within the next quarter.
Regulators will pay attention too. The Ministry of Road Transport and Highways has been consulting on an autonomous-vehicle testing framework since late 2024, and submission-grade simulation data has been a sticking point. A credible third-party simulator could short-circuit that debate. MeitY's draft AI advisory, currently being redrafted after the December 2025 stakeholder round, may also need to define how Indian street imagery is used for model training and who owns the derived simulations.
FAQ
When does the Street View feature go live globally?
Google has begun the rollout in the United States this week and has signalled a worldwide expansion over the coming weeks. Indian availability has not been confirmed by DeepMind, though historically AI Ultra features reach India within a quarter of the US launch.
How does this compare with OpenAI's Sora?
Sora generates passive video clips from a text prompt. Genie 3 produces an interactive world the user can move through, alter, and re-enter. The two are not direct substitutes, though both lean on heavy compute and may eventually converge into a single interactive-video stack.
What does this mean for Indian robotics startups?
It lowers the floor on simulation costs and pushes differentiation upward, into proprietary sensor data, Indian-specific edge cases, and road conditions that Street View does not yet cover.
Where can I read Google's original announcement?
Google has not released a standalone press note for this feature. The fullest write-up so far is TechCrunch's coverage from the I/O 2026 floor, linked in the source paragraph below.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.