How Does AI Learn to See in 3D and Understand Space?

Current AI models excel at 2D image analysis but lack a native understanding of 3D space, hindering applications in robotics and autonomous vehicles. A solution is emerging by combining three layers: metric depth estimation from single photos, foundation segmentation models like SAM, and geometric fusion. This pipeline uses AI to predict depth and segment objects in 2D images, which are then projected into a 3D model. The critical and most complex layer, geometric fusion, uses camera geometry to bridge these 2D predictions into a coherent, semantically labeled 3D scene, effectively creating spatial intelligence from flat images.