Physical AI
- In French:
- IA physique
Definition
AI systems that perceive, understand and act in the physical world: robots, autonomous vehicles, smart cameras. Training them requires varied, physically accurate data about the real world, which is slow and costly to collect, hence the use of simulation and synthetic data. The term is promoted in particular by NVIDIA.
Checked on against the references below. · Written with AI · checked against the sources Source: NVIDIA, glossary “What is Physical AI?”.
Concrete examples
Franka Robotics describes the FR3 Duo as a dual-arm system for physical AI research, bringing teleoperation, data collection and policy execution together on one platform. Its open-source LABS software converts recorded episodes into a LeRobot v2.1 dataset. NEURA Robotics says the 4NE1 Gen 4 runs on NEURON AI, its physical AI foundation model, and that its perception and memory functions run locally on the robot.
What it changes for you
A good simulated result does not guarantee real-world behavior: according to NVIDIA, simulated performance may not fully predict behavior after deployment. NVIDIA recommends comparing simulated results with real-world tests and addressing gaps before deployment.
Not to be confused with
NVIDIA contrasts physical AI with rigid automation: it says physical AI moves robots to autonomy that can generalize to new environments and tasks.
Additional material checked on .