Bridging the uncanny valley in CG
Lino Stephen Peter talks about his experiences in the uncanny valley after first becoming interested in the subject back in 2007 following the release of Beowolf.
Way back in November 2007 I started dwelling on the subject of the uncanny valley, having just seen the CG version of Beowulf.
The movie was well done, with all the latest technologies available today, but it evoked eeriness at some points; the animation and facial movement looked so awkward, that my animation experience and logic took over and I started analysing its shortcomings. It raised questions such as, why do all these characters look so creepy, and can’t we overcome this in this age of technology? This led me to the strange world of the so-called ‘uncanny valley’.
Masahiro Mori, a Japanese roboticist, coined this term in 1970 to characterise the relationship between humans and human-like robots. When humans interact with robots, they feel comfortable until a point where the degree of human likeness increases. Essentially, humans feel more comfortable with machine-like forms, and will get an eerie and strange feeling once they start to look more human.
This eerie feeling actually makes us feel more negatively towards the robot. This dip from the most comfortable feeling to a more eerie feeling he called the uncanny valley, bukimi no tani.
The graph in the image above shows the relationship between human acceptance of robots and their appearance.
Mori’s example of a prosthetic hand would seem humanlike from a distance until you started to interact with it. The Bunraku puppet creates some measure of unhealthy eerie feeling because of their strange movements and make-up.
This is seen more in animated subjects than in static ones. Thus Mori says as robots appear to look more human, they seem more familiar until a point where exaggerated imperfections create a sensation of strangeness. This is also the case with CG human characters that look creepy when designers try to push the envelope and add more and more realism to them.
CG realism has been used heavily since the dawn of 21st century, when the movie Final Fantasy raised the bar for humanlike CG. It didn’t do well at the box office, however, and many theories were put forward for the causes of failure of this movie – one of them being the unrealistic movements and eeriness seen in the characters.
Nevertheless, a group of people continued working on the technology to bridge the gap. The Polar Express and Beowulf were successful both in terms of technology and box office performance. Motion-captured acting also appeared in blockbuster movies such as the Lord of the Rings trilogy (Gollum in particular), the Pirates of the Caribbean trilogy (Davy Jones), and of course Avatar, where nearly everything was motion-captured.
Others have moved away from this phenomenon by opting to work with cartoon characters; others are struggling in this ocean to reach the shore; and still others are working to bridge the uncanny valley. I’m one of them.
Suspension of disbelief
When a human encounters a CG human character, they try to believe the character is real. A kind of suspension of disbelief takes place, but our subconscious mind is so primitive that it tells us that this character is not real by pinpointing microscopic problems with the character – and our believability goes straight down into the deep valley of eeriness.
This cognitive dissonance is a challenging phenomenon to overcome. Is uncanniness a thought or emotion? Both make sense since research has revealed that the uncanny valley happens because of improper movement, ugliness and weird behaviours.
When CG characters start moving, I believe they often evoke a human emotional reaction – the psychological terror that a counterpart looks like you. This is a terror management mechanism in human beings. Even if the characters look good, it’s difficult to digest the fact that CG characters are trying to replace you. This is also evident in day to day life; for example, in your job, when you meet an equal competitor for your skill you will immediately go into defensive mode.
This feeling differs from society to society. For example people from Japan are more acquainted with robots and comical counterparts. Their level of acceptance differs from a society whose acquaintance is less.
In the real world humans are well-versed in non-verbal messages from other humans, whether consciously or not. This subliminal expression is mostly missing in almost all CG character movies. Normally 70 per cent of what constitutes realism comes from getting it right in the face to the mid chest area; the rest of the body and clothing contributes around 30 per cent.
When one person speaks or makes a subliminal expression, we generally move our eyes between the face to mid chest area, and other movements go unnoticed. Since the face contributes most to realism, it’s very important to get it and the surrounding areas looking right. That includes believable hair movement, eye proportions, standard human face norms, textures, facial animation, lighting and so on.
Android scientist Karl F. MacDorman has conducted several studies in this area, including research on CG human faces, and has assessed what contributes to realism.
This was covered in four main studies. In Study I, texture photorealism and polygon count increased human likeness. In Study II, texture photorealism heightened the accuracy of human judgments of ideal facial proportions. In Study III, atypical facial proportions were shown to be more disturbing on photorealistic faces than on other faces. In Study IV, a mismatch in the size and texture of the eyes and face was especially prone to make a character eerie.
The CG artist
Using these results, he derived some conclusions for the CG artist. “The findings of this study suggest the following design principles,” he says. “To design attractive, human-looking faces that are not eerie, use high polygon counts with smoothing, and nearly ideal facial proportions. It may be safer to use a less photorealistic texture unless human photorealism is required. When using a human photorealistic texture, ensure the proportions of the CG face are within human norms. Finally, to prevent eeriness, avoid mismatches in the degree of human likeness of CG elements.”
Currently we cannot say that we have bridged the uncanny valley. In fact it is a continuous struggle – there are different aspects of human body motion that have to be reproduced in CG humans.
Studies in psychology and evolution of a hypothesis to design, develop and deploy CG humans will help implement this in various fields such as medicine, entertainment, the military and immersive technologies.
The uncanny valley can be bridged only when we develop a full CG character that is convincing when seen at any angle by the majority of humans.
About the author: Lino Stephen Peter is currently working on “Sultan the Warrior”, a full length 3D animation, responsible for pipeline setup, technical supervising, and technology innovation. Having an extensive experience in broadcasting and the media/entertainment domain, Lino trains professionals to suit the needs of the production.
on Thursday, December 30th, 2010 at 1:00 pm under Features, Opinion.
You can subscribe to comments.
You can leave a comment, or trackback from your own site.
Tags: Animation, Beowulf, CG expression, CG realism, Final Fantasy, Lino Stephen Peter, photorealistic, uncanny valley