Slashdot Mirror
Robots, Robots, Robots
destructor writes: "It looks as though robots can answer the questions of life and capitalism through robotic soccer simulations. I found this article over at megarad.com that tells us about Dr. Balch's experiments with soccer robots [NYT, free reg, blah blah]. For now, it is purely a computer simulation, but I guess it will be turned into a physical environment rather soon."
Additionally Shabazz writes: "The SF Weekly has a story about a band called 'Captured! by Robots' that started by Jay Vance (who some may recognize from Skankin' Pickle) and several robots that he created. The band is a bit out there, but something that any true geek can appreciate. Maybe this is the start of something great!" Additionally Phred noted that the Oregon Robotics Tournament being held this Saturday (December 1).
Jigsaws won't work, its too simple a problem (no teamwork required, no changing scenarios in real time), but I agree that something other than Soccer could be a better test of the bots.
Personally I think SKIING would be an excellent test of bots. And it would enable bot builders to justify funding for trips to the alps every now and again.
Ever changing scenario, quick responses, weight balancing, gates to get through in difficult visual conditions. Different disciplines - Jumping, Downhill, Boarding etc...
Admitedly balance element would be a limit at the moment - but given time this could be hugely entertaining stuff to watch when they lose their footing!
The general technique is to use an approximation of search (finding based upon luck) which tends to find good solution. Generally, it is based upon luck, and the only way to guarantee that its human-like is to restrict the search space to human-like possibilities. Whatever human-like means.
This is NOT the simulation of algorithms already created always, its often the creation of totally new algorithms, which can be extracted after development - a learning approach.
How? Straight statistics, support vector machines, decision trees, neural networks, fuzzy logic, and simulated annhealing are all common techniques to lead towards the goal. Who knows what they actually use.
Mod me down and I will become more powerful than you can possibly imagine!
A poster above had commented on Logical v. Illogical Actions. I'd agree with him, but what actually sets us apart is emotions, or instinct, however you may choose to call it since the two are subtly connected.
Robot emotions play a much larger part than most people know. In fact, any serious researcher into AI would know that emotions are nothing but another decisive factor, except that they are not well understood.
I'd seriously advice looking at Arthur T Murray's research into this area available here. It has a lot more to it than mere `entertainment` value.
And if you really are worried about robots having souls or a conscience, you should read this!. It is a pretty insightful article into what robots may have to do to qualify as humans.