NVIDIA researchers found Veeda AI: world models simulating physical reality for robot training

ArashVahdat · x · 2026-08-19

Sanja Fidler, Zgojcic and Huan Ling have founded Veeda AI, building multimodal foundation world models that "simulate physical reality" for Physical AI.

Their thesis: embodied intelligence will not emerge from imitation learning alone. Just as humans first observe the world and then build deeper understanding through interaction — and just as LLMs truly unlocked capabilities once trained on interactive tasks rather than passive internet text — robots' scaling moment will come through interactive learning. But real-world trial and error is too slow, expensive and unsafe, so we need simulated reality: environments where embodied agents can experience situations, make mistakes, and retry millions of times before touching the real world.

Veeda aims to build infinitely scalable interactive training environments to unlock the scaling moment for Physical AI.

Related event: Former NVIDIA Scientist Sanja Fidler Co-founds Physical AI Startup Veeda(6 posts)→

Original post →

More from Embodied

Embodied channel →