NVIDIA Opens Next Era of Physical AI Research With Agentic Technologies for Autonomous Driving, Robotics, and Vision AI

The core challenge of physical AI research goes beyond simply developing powerful models — it requires building an entire workflow that reconstructs real-world scenes, generates edge-case scenarios, and trains and evaluates policies. To address this, NVIDIA introduced NVIDIA Cosmos 3, the world's first omni-model unifying vision reasoning, world generation, and action generation, along with NVIDIA physical AI skills, which help researchers quickly turn model capabilities into scalable end-to-end workflows.

In the autonomous driving (AV) research field, NVIDIA offers the following technologies to tackle the 'long tail' problem where data collection is difficult:

• Neural Reconstruction: Converts fleet data into editable 3D scenes for use in simulation and synthetic data generation.
• InstantNuRec: Supports fast 3D Gaussian road scene reconstruction from images without per-scene optimization.
• NVIDIA AlpaGym: An open-source closed-loop reinforcement learning (RL) framework that scales across thousands of GPUs.
• NVIDIA OmniDreams: An action-conditioned generative world model that provides photorealistic rendering that responds in real time to policy actions.

Additionally, NVIDIA is further advancing autonomous driving research with NVIDIA Alpamayo 2 Super, a reasoning vision-language-action (VLA) model with 32 billion parameters.