When AI understands geometry, it can learn from raw video alone. This means no need for any sensor, label, or costly robot time. This kind of spatial awareness will change the world.
Take a look at this footage of a simple voxel world with two blocks that can't be fully seen from any single direction. Yet with an understanding of how this simulated world is built, we're able to reconstruct the entire scene.
Now imagine it has that same deep knowledge of the real world's physics and structures. What happens then?
Can't wait for what the future holds!