News StarVR, Tobii unveil next-gen StarVR One headset

"appeared to get cold fit"

A new form of workout, I presume?

 

InfiniteEye has come a long way from the early lets-stack-some-Fresnel-lenses-and-see-what-happens prototypes on MTBS3D.

 

"... claims Henrik Eskilsson, Tobii chief executive, of teh partnership..."

I want one. I want one so much.
And I sooooo can't afford one.

 

Not gonna lie, I love seeing "teh" in a professional publication. It validates my own excessive typo-ing of that word.

 

Looking through the product page the field of view comparison they claim vs other VR headsets strikes me as rather odd...
Are the VR headsets out in the wild actually as horrible as these guys claim?

(I'm talking about real VR, not some cardboard stuff for phones obviously)

 

Trying to represent field of view via a flat image is always going to be a failure, because there is no perfect mapping from a spherical view (or section of a spherical view) to a rectilinear (flat on a screen) view. There are a whole bunch of projections to choose from, and none will give a 'true' representation, just varying degrees of distortion.

On top of that, just "more FoV = more better" is not the whole story. If you take a panel and add two different optics to it - one with a lower field of view, one with a larger field of view - the view through that system with a wider field of view will have a lower pixel density (i.e. lower angular resolution). It will also have more distortion to the view (because the pupil physically translates as your eyeball rotates, the former off-axis you view the further from the true optical axis the pupil moves, and the further to the edge - or out of- the eyebox the pupil moves) but the presence of eye tracking should mean this can be compensated for by using a dynamic lens compensation function rather than static. You also start to hit rendering issues as you approach the 180° asymptote for rectilinear rendering (and much past 90° you start to 'waste' most rendered pixels around the periphery): you start to need to render multiple viewports for each view and then join these up to produce the final image, which results in rendered pixel density varying across the view.

Wide FoVs are much trickier than just 'use bigger lenses'. StarVR have the advantage that basically every application run on it will have been written specifically to run on it, but that doesn't mean it will be easy. Even if you had a dangerously bulging wallet and bought one to use as a substitute for a consumer HMD, you'd quickly find that you can't just stretch existing content to a wider view.

 

I wouldn't do that, but the addition of eye tracking is likely to spread to regular consumer VR (thinking Rift 2, Vive 2) because it would enable solving most of those problems you mentioned by dynamically adjusting the rendering to the reality of your eyes.. at least in theory (as it would obviously require software support and come with a performance penalty).

 

Foveated rendering requires far more than just software support. It's a massive rendering pipeline issue: you need to get from a camera capture of the eyeball to a future predicted gaze point (x2, for depth) then inject that into the engine as late as possible in order to minimise the time between render and display (no timewarping tricks here, because your goal is to not render those edge pixels). And unlike head movement where your motion-photons latency needs to be within 20ms, for eye movement (saccades) the latency needs to be single-digit ms at most.
Even the camera hardware for ultra-low-latency eye tracking for real-time display are not available even in academia (thus far no call for it) let alone off-the-shelf. Everyone working on it is building their own hardware, which thus far requires FPGS (will later by ASICs) sitting right on the back of the sensor itself to do the processing.

 

I've been watching VR for a while now. I figure by the time I have a decent job, consumer VR headsets will be good enough to buy with either a single wire or wireless.