engineer

tim   .

                                                      

.   hunkin

cartoonist

THE UNCANNY VALLEY

  I wrote this for a fundraising event in December 2009. Its a short history of the attempts to make lifelike robotics and animatronics. If one of the you tube links stop working, please let me know.

    When we look at the things around us, we can’t help trying to see everything as a face. The protractor is obviously a happy protractor and the mole grips look hungry. Three dots are enough to make a face, it’s the first pattern babies recognise.

. .

The reason we see faces everywhere is that our brains are particularly sensitive to them. Our eyes are really quite crude sensors, its our brains that put the limited information together to make a convincing perception of the world around us. The basis of most optical illusions is the mismatch between the limited information coming from the retina with the huge assumptions made by the brain.

The reason we see faces everywhere is that our brains are particularly sensitive to them. In our evolution its been very useful to be able to recognize individuals, and be sensitive to their mood. This is what makes us ‘see’ faces in everything. Even people who say they aren’t observant effortlessly recognize their friends and judge their moods. But our sensitivity to faces also makes us particularly observant at spotting subtle differences between a real face and a waxwork or robot. This talk is about the quest to make a human replica completely convincing.

The title of my talk comes from Masahiro Mori, who was professor of robotics at Tokyo university in the 1970s. He produced a graph he called the uncanny valley. His idea is that industrial robots don’t look very human and so aren’t seen as friendly or possessing any emotion. Making robots appear more human, we can see them as more friendly. Make them too lifelike though and they start to be creepy. At the pit of his graph, the creepiest thing imaginable, he places a moving real dead body. (He came up with his graph after the shock of shaking hands with someone with a prosthetic arm).  Halfway up the slope up towards totally accepting reality he puts ill people, handicapped people and Bunkaru puppets. It’s over simple, because people vary enormously in what they find creepy.

I’m partly interested because a lot of people find my arcade in southwold creepy, including my grandson. My chiropodist was inspired by a strict nurse with a brisk efficient manner, but I overheard some describing her as a zombie last week. I still don’t see her as uncanny, though I know others do. The uncanny valley graph has its limitations but its a great title and I think the quest to create a convincing human replica is a great subject. 

  . .

On the left is the Dagenham idol, the earliest known depiction of a human ever found in Britain. Not at all lifelike, but our brains completely accept that it was intended to be a human figure. The first step towards the uncanny valley was to get the overall shape right. I think it was the ancient Greeks who first did this – this is partly why we admire their statues so much. They were interested in an abstract idea of beauty. However, the Romans were the first to capture lifelike ‘portraits’ of specific individuals (see right). Their portraits were originally painted, so they would have looked a lot more ‘real’ than they do now.  

Progress is complicated because in the middle ages, people weren’t generally interested in realism. But it came back with the Renaissance, this16th century Spanish Christ is made of covered in gesso with ivory teeth and glass eyes is an example.

But stone and wood can never look really convincingly like skin. Skin is translucent, so some of the colour comes from below the surface. It also has hairs and a fine surface texture. A better material for copying skin is wax. By the 19^th century ‘waxworks’, which people paid to visit, had become popular. The appeal was the spooky realism of the figures, some of them death masks of criminals. Its amazing that they still thrive today because of our obsession with celebrity.  I love the publicity photos Madame Tussauds, London, take of the celebrities standing by their waxwork. You can always tell which is real.

  .

I think the convincingness of waxworks has more to do with how they are viewed than how beautifully they are made. Madame Tussaud’s has always had a wax security guard sitting by the door – I enjoy this more than the celebs because its unexpected. The ultra realistic figures made by the artist Duane Hanson are effective because they merge with the gallery visitors.

Wax remained the most lifelike material for skin until the 1960s, when silicon resins became available. A big advantage is that silicon can be moulded from real skin to faithfully reproduce its fine texture and also that its easy to thread fine hairs into silicon. The artist Ron Muek uses silicon for many of his figures. Ron is a great craftsman and his body shapes, faces and skin texture are often completely convincing. Dead Dad, who is half size, is certainly down there near the bottom of the uncanny valley. But for me, he’s not quite at the bottom. Ron’s figures, though they look so real, never move…... moving replicas are much more creepy.

Making a moving realistic realistic figure is much harder. Their are many myths about them. The golem is a jewish giant with many stories attached to him, always fashioned from clay. The giant Talus, made of brass, is said to have guarded Crete against intruders by heating up his body and hugging them to death. His only vulnerable spot was his right ankle, which had a sinew of flesh and a vein of blood. There’s a bone in the ankle still named the Talus.

Though creating lifelike movement is hard, people obviously have an innate fascination because over the centuries an enormous amount of much effort has gone into it.  It was easier to start small. The first practical moving figures were possibly Heron’s simple ancient Greek automata, powered by water. Next were the medieval clockmakers, using clockwork to power complicated automata. They were made from the 17^th century onwards as one offs for the extremely rich. By the 19^th century France had quite an industry making clockwork automata. The video is rather slow but it gets increasingly creepy. The automata were always small, so they weren’t quite as creepy as you might expect.  I don’t really understand why clockwork can’t be scaled up, though it obviously can’t be as all clockwork mechanisms are small.

The goal of a full size human replica remained remote until people started playing with electricity in the 18^th century. As they got electric shocks, they quickly realized that we are affected by the stuff. But the key experiment linking our bodies to electricity was done by Galvani. His classic experiment was to make frogs’ legs twitch (see all the frogs legs in his lab, above).  I later found that he also tried it on chickens and dead human bodies. Reports of his experiments are partly what inspired Mary Shelley to write Frankenstein.

 At the time, no one knew enough about electricity to put it to any practical use, but it certainly made the idea of replica humans more ‘high profile’. Depictions of steam powered men and servants became common.

100 years later small motors and relay logic (to switch combinations of the motors on and off) made it finally possible to construct a crude mechanical man. Electro the smoking robot was inspired by the Czech play RUR, which coined the word ‘robot’. Robots quickly gained popularity, and began to be made as toys, to be collected, and to star in films and comic strips.

There were still formidable problems to make anything remotely realistic. The first problem is the jerkiness of the movements. When we move our arms several different muscles stretch or contract and they smoothly work together at constantly varying speeds. The ‘Robotic’ jerky movement of early robots that are still performed by mime artists come from the motors and pistons, which naturally move at constant speeds.

Computers have helped to smooth the motion. Industrial robots now have sensors in all their joints to provide feedback to the computer and this controls the speeds and accelerations of the motors. This gives them fluid multi axis movements which can be very balletic. Though the movement looks effortless, robot arms require a lot of precision parts and fast computing, so one like this costs at least  £100,000. They are now so reliable that people trust them with their life.

Practical robots are no longer mechanical men, they have become completely functional. Robot researchers now tend to be more interested in mimicking animals than humans. Many legs is easier than just two.

Its no problem getting a robot to walk if it has something solid to hold to "support" it. This robot man was a seaside attraction in the 1950s.  Without a prop like his carriage, walking is hard. We walk without thinking, but keeping in balance while moving the top heavy weight of our head and body is actually an amazing feat. As we walk we are subconsciously and constantly adjusting our balance. Robots have great trouble reacting as fast as we can. As far as I know, no robot can catch a ball. A japanese team has been trying to get a robot to juggle, but so far it can’t react fast enough.  This is a robot made by Honda for research. It cost millions. 

Most recent attempts to make humanoid robots don't attempt walking because its so hard. One of the reasons I’m interested in the uncanny valley is that  a fiend of mine called Will Jackson has been developing a robot for the last 5 years. Called Robothespian, it is mainly sold to science centres to greet visitors and perform in their Object theatres. It’s engaging, particularly because of its eyes, which are two mobile phone display screens each showing video of an eye.  I think its quite friendly, but I’m sure some people find it very uncanny.

Unlike most robots, Robothespian's limbs are ’air muscles’. These are rubber tubes which simultaneously swell in girth and contract in length when fed with compressed air. The advantage is that unlike ordinary pneumatic cylinders, the amount of movement is proportional to the pressure applied.  A computer moves and positions every air muscle using proportional pressure air valves. The drawback is that the movements aren’t precise. Robothespian currently has trouble getting fingers of both hands to touch, though its really impressive how Will is continually improving it. Will’s struggles have made me aware of just how much I take for granted about the seemingly effortless way my body moves about.

Even if you get the movement right, to make something really lifelike there’s the skin problem. With old automata, the leather and fabric quickly cracked up because of the movement. Today silicon rubber can stretch to some extent, but it needs to be very thin to stretch as much as skin. If its thin, it doesn’t keep its shape and also tends to tear. There's no perfect plastic for skin, so animatronic figures usually have rather limited facial movements.

Disney imaginers have been making "realistic" animatronic figures since the 1960s. Disneyworld, Florida, has a hall of presidents, each of which gesticulates and says something in turn. They have PVC skin, with stage make-up on top. The necks unzip at the back to get at the mechanisms inside. From the front they look convincing, but they are seen in low light and seen at some distance. To move the arms, they use hydraulic pistons, miniature versions of the ones on diggers. Hydraulics tend to leak oil, so you never see the platform the figures are standing on. Personally, I found the seriousness of the presidents presentation made them more funny than creepy.

Disney is secretive about their technology – they have never patented anything for fear of having to expose their secrets in the patent application. However they have competitors who are quite happy to let people take photos. The Sally corporation makes anything from Greek goddesses to the undead. The Sarcos corporation does the same. Although Sarcos primarily makes animatronics for theme parks, it also makes sophisticated sensors and prosthetic arms. Animatronics also completely overlap with research. 

The most creepy of the animatronics companies is the Hanson Corporation. Their technology is based on their skin material they called flubber. Jules is their creation. They deliberately didn’t add a back to his head just so you know he isn’t real. He looks as if he’s had too much botox to me.

However, technology keeps moving on and Darpa, the American military research funding organisation that came up with the internet and the computer mouse had been busy funding walking robots. This is Big Dog, made by Boston Dynamics, founded by the genius Mark Railbert, who also started MITs leglab. To me, though this doesn’t look remotely human, it is really uncanny, right at the bottom of the valley. Amazingly, though the basic leg movements are high pressure hydraullics, the reactions when it falls come mainly from huge springs, not some high tech feedback system. 
 

home

search

about the site

site
map

links to
other sites

where to see stuff

contact me