Androids at China's Robot Expo 80
eldavojohn writes "China's 2006 Robot Expo has wrapped up. Even though there is little information on it online, there has been much attention given to Zou Renti's android. It seems that everyone cool is making androids of themselves these days. There's a decent article on the state of androids in Japan but unfortunately, the concentration isn't on functionality, it's on fooling the humans the robot interacts with: "The key to a successful android, according to Dr. Ishiguro, is both very humanlike appearance and behaviour. One of his early android creations was cast from his then four-year-old daughter. While it looked like her, it had few actuators and its dull facial expressions and jerky movements proved so uncanny that the girl later refused to go to her father's lab because her scary robot double was lurking there." The latest robot he's built has 42 actuators, allowing it to wow many spectators at the expo. I wonder how much longer it will be before we see Blade-Runner-like cases on the evening news?"
Forget these androids (Score:4, Interesting)
Regardless, these androids are carnival mannequins with better fake skin. They are also victims of the "Uncanny Valley [wikipedia.org]". At worst they look cheap, at best they're creepy. I got a picture taken with one [mcgill.ca]. The developers refer to it as a "lover robot" and it would move its mouth while piping a Celine Dion song through a speaker. They spent way too much time adding fake nipples and revealing clothing.
The product brochure by the "Beijing Yuanda Superman Robot Science A Company of Limited Liability [bjydcr.com]" states:
"The lover robots like the real beautiful woman and handsome guy are primarily for family collection and appreciation. This is a huge market, for instance, recently Japanese will spend about 27 billion yuan on person-like robots each year, and the global consumption on such commodity is about 500 billion yuan. Comparing with these unmovable puppets, the lover robots are more realistic, charming, intimate, lovely, sexy and attractive."
Re:The key to a successful android... (Score:5, Interesting)
Look at Japanese humanoid work (Score:5, Interesting)
The important site to look at is Robots Dreams [robots-dreams.com], which covers Japanese robotics work. The little humanoid robots at the $1000 level are getting quite good mechanically. The best ones now have maybe 70% of the hardware functionality of Asimo at under 1% of the cost. They're typically remote controlled, but, because they have more degrees of freedom than a human can control with an R/C controller, preprogrammed movements were added. That wasn't good enough, so some hobbyists have added gyros and balance reflexes. Now it starts to get serious.
The hobbyists are doing some very good work. There are competitions and battles for these things. Obstacle courses which look like something from Army basic training. The battles aren't just banging away like Robot Wars; these machines can execute judo throws.
More to the point, the hobbyists are making progress much faster than the academic robotics people ever did. There are more of them, enough to drive a market for mass-produced parts. That makes it easier to build the things.
If you took the best kit humanoid available (which costs about $1200), added a 6DOF inertial unit (a few hundred dollars and getting cheaper every year), a stereo vision system (or even a SwissRanger [swissranger.ch] minature LIDAR), better force sensors in the feet and hands, and a WiFi link, you'd have ASIMO-level capability for a few thousand dollars. We'll probably see that within two years, and probably at a low price point.
Then it's all software. And there's lots of theory out there just waiting to be used. This is going to be fun.
Re:The key to a successful android... (Score:3, Interesting)
OK, I know where people get this, and this sort of thing keeps flowing around, so maybe I can help, just this once.
IAAAIR. (I am an artificial intelligence researcher).
First: neural networks do not, in the general sense, run programs. They get trained to execute what basically amounts to mathematical functions. One function-crunch per cycle, roughly. (I'm vastly simplifying here, but the main point is that neural networks are not, as currently implemented, general-purpose computing devices).
The brain is not a neural network in the generic sense used by programmers. Neurons don't operate that way. And it violates a definition of algorithmic behaviour by not having a clear `beginning' and `end' state of any given operation. At best you can close your eyes and pretend for a while that it's a collection of disparate communicating self-modifying neural nets. Whoops, you just left out brain-wide effects like stress hormones, nutrition and so on, even ignoring hallucinogenic crap people feed themselves for kicks.
Quantum computing, too, does not allow escape from the basic limitations of turing machines. There's some discussion about whether it equates to a nondeterministic turing machine, but it looks like it doesn't even do that.
Even so, let's say that you build a neural network which does The Right Thing(tm); how do you know it's intelligent? Answer: you really don't. We don't have any clear idea of how to reverse engineer the operations of neural networks. So maybe it's intelligent (in the sense of having automated non-human cognitive processes) or maybe it's just a convincing act. You don't know, and until you learn to analyse neural networks (be sure to tell the Nobel folks) you won't know.
If we're going to build an AI (I hate that phrase), and prove that we have done so (contentious point number 5239 in the field), we have to understand the operation of what we assembled. This means we need to have a definition of cognition, an implementation plan which covers the necessary subcomponents and communications between them, and some way of monitoring and interpreting that behaviour. Some components may actually be neural networks. Some may not. My money says that we'll have neural nets doing jobs they're good at, such as pattern recognition for perceptual tasks. At that level one hardly cares how the Android of the Future sees the Klingons, one just wants the identification to be reliable enough so that he doesn't shoot Picard.
By the way, if you're really, really driven to get involved with something like this, and are tired of having a life (or you're tired of your favourite realdoll) then reply to this message. If not, well, this world has plenty of plastic pundits, what's one more?
Re:Forget these androids (Score:3, Interesting)
I don't think you CAN spend too much time on nipples and revealing clothing.
And I'm not kidding - you want to make high-brow robots? Make and sell sexbots first.
Once you're a billionaire, you can make that chess partner and Jupiter Probe Pilot.
Science history will remember you for the AI work, but sexbots will pay for it.
Who wants to take bets on the first lawsuit over a celebrity replica sexbot?