Fei-Fei Li’s World Labs debuts Atlas, a world model showcase for advanced spatial intelligence
World Labs Inc., the high-profile and well-funded artificial intelligence startup co-founded by the renowned computer vision pioneer Fei-Fei Li, has just dropped Atlas, which promises to be a game-changer in the world of “world models.” In a blog post, World Labs explained that Atlas is a breakthrough multimodal world model that aims to bridge the […] The post Fei-Fei Li’s World Labs debuts…
Fei-Fei Li, a renowned computer vision pioneer and Stanford University professor, co-founded World Labs Inc. in February 2024 with the goal of developing advanced artificial intelligence systems capable of understanding and interacting with the physical world. The company's latest creation, Atlas, is a multimodal world model that represents a significant leap forward in the field of "world models."
Atlas is designed to generate detailed simulated 3D environments from single image inputs, with precise camera control, allowing for viewing from any angle. This breakthrough technology is built on a complex multimodal autoregressive diffusion transformer architecture, which exceeds the capabilities of earlier video generators.
Atlas excels in creating expansive, highly-detailed simulated environments, enabling users to view them from any perspective. The model is capable of generating up to a minute of 1440p video while maintaining geometric consistency and can output 3D assets such as point clouds and 3D Gaussian splats. This allows for the combination of multiple data types, including video, text, camera poses, and depth maps, to create a shared spatial context.
World Labs has demonstrated Atlas's potential in various applications, such as creative visual effects, game design, and robotics training.
One of the most promising aspects of Atlas is its ability to support "c" workflows, where developers can capture a physical space using an ordinary smartphone camera and reconstruct it as a 3D simulation that robots can navigate within. Atlas can generate precise RGB images and depth sensor readings as robots move within the simulation, creating a realistic environment for testing hardware and software.
In benchmark tests against industry-leading video generation and 3D reconstruction models, Atlas has demonstrated superior performance in terms of camera-path adherence and 3D geometry reconstruction from sparse inputs.
While Atlas faces stiff competition from other AI startups in the world model space, its unique ability to unify camera control, video generation, and 3D reconstruction sets it apart. The real test for Atlas will be in its wider availability and performance in real-world scenarios. Currently, early access is available for select enterprises, but no official release date has been announced. The success of Atlas will depend on its ability to replicate its impressive results in complex, real-world settings.
Written by urgent.news from SiliconANGLE's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
