Ant LingBot Open-sources LingBot-Map: Real-time 3D Reconstruction with Ordinary RGB Camera

Ant LingBot has officially open-sourced the streaming 3D reconstruction model LingBot-Map. Using just an ordinary RGB camera, it enables machines to perform real-time camera pose estimation and scene 3D structure reconstruction during video capture, filling key technical gaps in real-time spatial perception and providing reliable spatial understanding for robots and autonomous driving.

Unlike traditional "capture first, process later" models, LingBot-Map uses autoregressive modeling and geometric context Transformer to process frames sequentially, balancing reconstruction quality and efficiency with 20FPS real-time inference and stable long-sequence performance.

It excels in benchmarks: Oxford Spires dataset shows an absolute trajectory error of only 6.42 meters (2.8x better than previous streaming methods); ETH3D benchmark achieves an F1 score of 85.70 (8% higher than the second place). Ant has also open-sourced models like LingBot-Depth and VLA to strengthen embodied intelligence foundations.