Physical AI

AI that can perceive, understand, and act in the physical world — robots and self-driving cars are examples.

Physical AI is a term heavily promoted by NVIDIA, officially defined as systems that can perceive, understand, and reason within the physical world, then execute or coordinate complex actions. It covers robots, self-driving cars, and “smart spaces” that use fixed cameras to optimize factories and warehouses. The problems it addresses largely overlap with what academics usually call “embodied AI,” but its scope is wider — it even counts bodiless “smart spaces” that rely only on fixed cameras, with no robot body at all. NVIDIA lays out a three-step development pipeline: train models in a data center; simulate and generate synthetic data on the Omniverse simulation platform and with the Cosmos world foundation model; then deploy to edge computing platforms such as Jetson and DRIVE. When Cosmos was announced at CES in January 2025, Jensen Huang said “the ChatGPT moment for robotics is coming.”

ExampleA self-driving car processing sensor data in real time to decide steering and braking, and an autonomous mobile robot in a warehouse dodging obstacles while moving goods, both count as what NVIDIA calls physical AI.

Also called

Generative Physical AI

Related

Embodied AI · World Foundation Model · NVIDIA Cosmos · NVIDIA Three-Computer Solution · Sim-to-Real Transfer · ChatGPT Moment for Robotics

Sources

What is Physical AI? (NVIDIA Glossary)
NVIDIA Launches Cosmos World Foundation Model Platform to Accelerate Physical AI Development (NVIDIA Newsroom, 2025-01-06)

As of

2026-09