It's definitely not positionally tracked, looks like it's just a nice imu + touchpad.
Positional tracking is a really challenging problem with a lot of limitations. If they were using computer vision to track the controller it would need to be held within the device's FOV and have identifiable elements such as a marker or LEDs. There are other approaches, but nothing I've seen that would work well on an HMD.
I've been working with AR/VR for the better part of a decade, and controller tracking's come up several times. I can tell from my own work that the device they're showing off is IMU only. It's too bad, something equivalent to the Vive controllers for mobile VR would be fantastic.
And at the end of the day - you're right. If it was spatially tracked then they definitely would have mentioned it.
First, I think positional tracking of the person within his/her environment is one of the main things that makes the difference between VR and AR. So there's value in that. If you can do it with computer vision maybe it can work in almost any environment.
Second, there are RF based positional tracking systems. They're more expensive, true (not if you can design chips yourself, so might even be cheaper for google), but you don't need the controller to have visible elements.
Third, hand tracking like this should be able to work usefully, shouldn't it ? I know it doesn't do distance, but ... Just hang it off the viewer pointing downwards with a large volume immediately in front of the user. Not great for gaming probably, but for a virtual keyboard it should work, no ? http://www.ctxtechnologies.com/products/vk-200-keyfob-virtua...
There'd need to be either:
- a very expensive IMU in the controller, talking $300-500
- a camera and hefty CPU in the controller,
- a camera and dedicated image processing CPU on the headset
To be able to integrate the acceleration data from an IMU to get positional data out of it is not something you can do with commodity IMUs, even the one made for Oculus. They're too imprecise and too inaccurate and too slow, and any analysis tricks to improve the results all introduce latency. Oculus' sensor has very little drift (though it still has drift, which is another thing the camera system on the desktop Rift can correct for) and the noise in the orientation data is not noticeable or can be filtered much more simply without introducing "too much" latency.
There are IMUs that are good enough to integrate acceleration data to position data, but they are prohibitively expensive for this application.
For what it's worth, I used to see a marked difference in the quality of orientation tracking between my Galaxy Note 4 in Google Cardboard vs. the Gear VR, but I see no different between GC and GVR on my Galaxy S7.