AI developers shift focus from text to 3D world models
AI is evolving to simulate real-world physics for training robots and autonomous vehicles in high-fidelity virtual environments.
(Photo via Wired)
Artificial intelligence is moving beyond text-based processing as researchers shift their focus toward 3D world models. According to a report published by Wired on Nov. 8, 2026, this technology is designed to simulate real-world physics to provide training environments for robots and autonomous vehicles.
The transition toward what is described as Physical AI requires systems to gain a deeper understanding of geometry, dynamics, and semantics. By operating within high-fidelity simulations, these models allow machines to practice navigating reality without the limitations of slow or costly real-world testing.
These simulated environments enable robots and autonomous cars to experience rare or dangerous scenarios, such as extreme weather or complex traffic encounters, that are difficult to capture at scale in the physical world. This approach aims to accelerate the development of systems capable of interacting safely with their surroundings.
The industry has seen rapid movement in this sector recently. On Nov. 6, 2026, NVIDIA released four world models as part of its Cosmos 3 family, specifically targeting robotics and autonomous driving. Other entities, including Waymo, have also contributed to the field with the introduction of the Waymo World Model in February 2026, alongside ongoing research into projects like PAGE-4D.