Nvidia Tackles the Uncanny Valley with New Rendering Tech
The Uncanny Valley is proof that sometimes doing something imperfectly is tacitly worse than not attempting it in the first place. For the uninitiated, this is a term that's been floating around the field of robotics and computer animation for quite some time. It's a theory which holds that when something looks and acts nearly human - with a few small imperfections - the automatic response on the part of any human observer will be inevitably negative.
The "Valley" itself is said to be a negative dip in emotional response right before the point at which something looks and acts fully human. Depending both on the observer on where in the 'valley' a particular subject falls, the response might vary. Perhaps the observer will only feel slightly unnerved; a vague sense of 'wrongness' may creep into their mind without them fully realizing why. On the other hand, they might immediately be agitated- even terrified - at the sight of the not-quite-human subject.
It's difficult to say how much weight the concept actually holds; potential reasons behind this fear response range from mate selection to a fear of death to conflicting visual and perceptual cues to survival instinct. At this point, I'd request that you all try a little experiment: pull up a picture of Charles Manson (or really, any sufficiently deranged serial killer) and study their eyes. For many of you, this might well produce an effect not dissimilar to the image I've posted below.
Most you probably don't feel particularly comfortable staring at the above image, do you? In that regard, the theory of the Uncanny Valley definitely seems legitimate. Why else would these images bother us to such an extent? How else can our agitation be explained?
Though the Valley is of some concern in the field of robotics, it's equally as pressing in gaming, though for different reasons. See, anyone who's played a game knows that immersion is important - particularly in a narratively-driven experience. Consequently, when you're looking at a face which troubles or outright frightens you - that is to say, when you're forcibly driven down into the depths of the Uncanny Valley -it's difficult to retain that sense of immersion.
Getting faces to pass our subconscious cue tests - getting characters out of the Uncanny Valley - has historically been one of the most difficult hurdles facing any graphic designer. For the past several years, they've been doggedly chipping away at the problem; determinedly crafting a staircase to take them up onto the valley's other side. Slowly but surely, they're making progress.
The latest development in the field of graphic design comes from Nvidia. Known as "Digital Ira," Nvidia claims the technology "represents a big leap forward in capturing and rendering human facial expressions in real time, and gives us a glimpse of the realism we can look forward to in our favorite game characters."
According to Nvidia, Ira is the bleeding edge of performance capture technology. Each and every expression on Ira's face was acted out in a "light stage" at the USC Institute for Creative Technology. There, a team headed by Paul Debevec photographically captured the facial features of its actor. Even more exciting is the fact that, unlike with traditional performance capture technology, no face marking or special make-up was required. The information is then fed into Nvidia's new engine, which uses something known as FaceWorks to render everything with rather startling detail.
To my eyes, Ira still doesn't look quite human...but he's definitely an improvement over some of the stuff we've been seeing.
An interactive demo is currently live on Nvidia's website, which allows observers to view Ira in three different lighting environments and he moves through a range of realistic expressions. Those of you who are interested can grab it here.