Synthetic Data Generation

AVES Reality provides the spatial foundation for NVIDIA’s Smart City AI Blueprint. With AVES Launcher, geospatial data is transformed into accurate, georeferenced 3D digital twins with HD road maps that reflect real streets, junctions and city districts. These city models are simulation-ready and integrate with platforms such as NVIDIA Omniverse and CARLA, positioning AVES as the entry point of the Cosmos Transfer workflow.

Within this blueprint, the AVES digital twin is used to define where and how virtual cameras move through the city and how traffic behaves. On top of the AVES world, driving scenarios are simulated and multi-sensor control videos are generated, along with precise annotations such as depth, segmentation or object information. This creates a structured, repeatable way to produce training data that is directly tied to real urban layouts.

Cosmos Transfer then uses these AVES-based simulations to generate photorealistic video variations – for different weather, lighting, viewpoints or traffic conditions – while preserving the underlying city geometry. Because everything starts from an AVES digital twin, the resulting datasets are not generic: they are grounded in specific cities and can be aligned with real camera positions and use cases. In this way, AVES is at the centre of the Smart City AI Blueprint, enabling partners to rapidly move from raw geodata to high-quality, city-scale training data for perception, analytics and planning.