I think we're going in that direction, but current SLAM techniques aren't fast enough to provide the extremely low latency needed. John Carmack has mentioned prototypes that mostly work as long as you have dedicated room markings, but it gets unreliable when the environment can't be controlled, so it's a long way from consumer-ready.

If it could be done now, JC would definitely be getting it into the next GearVR. The other problem is that even with current best-case everything, something like GearVR doesn't have enough processing power and if it did, it would kill the battery.

The Vive already has a front-facing camera that it uses for Chaperone and a few other things, but it's not involved in tracking at all. In the short-term, expect other HMDs to borrow this idea.