XNews

Spatial Intelligence and 3D Scene Reconstruction

Spatial Intelligence and 3D Scene Reconstruction

Fei-Fei Li's World Labs announced the integration of SceniX, a spatial interaction company, to advance spatial intelligence beyond perception to interaction. Simultaneously, a new open-source model was released that can reconstruct 3D scenes from monocular video in real time, running at 20fps on a single GPU, stable over 10,000 frames without drift, and tested on drone, driving, and indoor footage. The model requires no LiDAR and outperforms traditional optimization-based methods.

Key insights

World Labs' SceniX Merger Targets Interactive Spatial AI

By absorbing SceniX, World Labs signals a strategic move from creating 3D representations to enabling AI agents to interact with those worlds. This aligns with the broader vision that spatial intelligence must encompass both perception and action, moving toward robotics and AR applications.

Open-Source Real-Time 3D Reconstruction Democratizes Spatial Perception

The new model eliminates the need for expensive LiDAR sensors and runs on commodity hardware, making high-quality, persistent 3D mapping accessible for drones, autonomous vehicles, and VR/AR. Its 10,000-frame stability and speed suggest practical deployment in dynamic environments.

Source posts