Skip to content

Dataset · nvidia

NVIDIA autonomous driving synthetic camera videos

nvidia/PhysicalAI-WorldModel-Synthetic-Autonomous-Driving-Scenarios

Synthetic panoramic traffic video clips with VLM descriptions for training world models and autopilot.

What's inside

The dataset includes between 100 thousand and 1 million temporally consistent traffic video clips generated in the NVIDIA Omniverse simulator. Each clip represents a 360-degree view from multiple cameras on an autonomous vehicle and is accompanied by English text descriptions generated by VLM models for each camera. The dataset was created to fill rare and complex long-tail scenarios missing from real driving logs.

What you can do with it

  • Training generative world models for traffic scenario simulation
  • Fine-tuning multimodal models and autonomous driving perception systems
Languages
English
Size
100K–1M rows
Downloads this month
71.4K
Likes
44

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/PhysicalAI-WorldModel-Synthetic-Autonomous-Driving-Scenarios")