NVIDIA Physical AI Day: Betting on Simulation and OpenUSD

At NVIDIA Physical AI Day, the company systematically showcased its latest layout in Physical AI and embodied intelligence. The event sent a clear core signal: the development of Physical AI relies heavily on simulation technology, and its underlying infrastructure is being redefined.

Standards and Foundation: OpenUSD and Synthetic Data

NVIDIA is aggressively pushing OpenUSD to become the universal foundational standard for 3D content, simulation, and Physical AI workflows, urging widespread industry adoption while promoting the Alliance for OpenUSD. Officials emphasized that Physical AI starts with simulation, with OpenUSD and synthetic data forming a solid foundation for this technical path. Additionally, NVIDIA is promoting free OpenUSD-related resources to lower the barrier to entry.

Shift in Data Generation Paradigms

The training paradigm for embodied and Physical AI is transitioning from "slow physical data collection" to "scalable, generalizable data generation." NVIDIA proposed leveraging world foundation models to drive data generation for Physical AI, solving the bottleneck of traditional data acquisition.

Simulation Moves to Real-World Applications

In terms of specific applications, Lightwheel AI showcased a practical embodied robot case study at the event: the robot first completed simulation training in the Isaac Lab Arena, and was then successfully deployed to a real operating room to perform relevant tasks. This marks the accelerated implementation of simulation technology in complex scenarios like healthcare.

2026-07-22 ~ 2026-07-23 · 6 related posts