Fei-Fei Li's WorldLabs Unveils Atlas: A World Model for 3D Space
创业邦 · wechat · 2026-09-02
Fei-Fei Li's WorldLabs launched Atlas, a 'spatial intelligence' world model. Key features:
- Camera Control: Generates videos with precise, pixel-level camera trajectories from 1-6 photos, outperforming competitors in blind tests.
- Sparse View Reconstruction: Reconstructs detailed 3D scenes from 2-25 photos using Gaussian Splatting, outputting depth maps/point clouds.
- Robotic Training (Real-to-Sim): Rebuilds real-world environments as simulators for robots, generating infinite training variations (R2S2R pipeline). It aims to bridge the gap between 'seeing' and 'understanding' 3D space.
Related event: World Labs Unveils Atlas, an Omni World Model for Spatial Intelligence(47 posts)→
More from Embodied
- Asimov 1 Robot Update Simplifies Assembly, Open Sources CAD Files — max_paperclips · 2026-09-02
- Gravity-Prior Driven Decoupling for Robust Pose Estimation — zhenjun_zhao · 2026-09-02
- Diverging Autonomy: Waymo Hits 500K Weekly Rides While China Deploys 3,800+ Driverless Mining Trucks — 创业邦 · 2026-09-02
- Expert: 1 million humanoids in US jobs within a decade, maybe — binarybits · 2026-09-02
- Digimon AI Project: Reward Hacking and Progress in PPO Training — redfoxkiller · 2026-09-02
- ZimaBlue: Evolving Generalizable World Action Models via Video Pre-training — JoyFutureAcademy · 2026-09-02