If I had to make a guess as to it's use..
One would be to dynamically construct game 3D environments for use in both AR and VR. In the later case it could help VR users keep their spatial awareness without breaking the immersion of a virtual world. For the former it's a rather necessary step in superimposing graphics onto the real world.
There are some technical hurdles I'm not sure if they've sorted out. One being pathfinding for AI in VR. When a 3D landscape can be created in advance, pathing can be tested as well. I imagine procedurally generated worlds face some AI problems as well but the real world is harder since it doesn't follow an easily definable set of rules. Though some of the tech from self driving vehicles could be applied here. Which google obviously has experience with.
Though I suppose Tango could also be used to crowed source the generation of 3D models of.. well the entire earth given enough time. A 1:1 scale replica of the entire earth updated in near realtime? Kinda scary as much as it is neat.