Here's how we're building a fleet of digital twins for our autonomous vehicles — and teaching machines to perceive the world across diverse conditions.
What it is
The photorealistic simulator generates virtual driving scenarios, including those based on real-world ride captures. Using Generative AI, we can modify and recreate new conditions and scenes. The output is then used to train our vehicles for full autonomy — and transfer that learning to real-world conditions for Physical AI transformation.
How it works
Imagine exploring dynamic street panoramas on a map that you can freely edit — add objects, change camera angles, adjust weather, or tweak any other parameter. And the world around reacts in real time: shadows shift, textures update, depth and distances to objects adjust.
That's exactly how our photorealistic simulator works — we can generate countless variations of new situations in any environment by combining generative models with our proprietary neural networks.
Why we built it
With the photorealistic simulator, we can generate hundreds of thousands of kilometers of new driving scenarios — without expanding our fleet or running millions of physical miles. This allows us to train the advanced AI models that power our autonomous driving technology. High-fidelity simulation testing guarantees reliability and safety.
Where it leads
Ultimately, a system trained on simulations will help us replicate any real-world driving scenario under any conditions — bringing us one step closer to full Level 5 autonomy.