Posted by
emmett
on from the more-expensive-robots dept.
adamy writes "Seems that a Japanese firm is selling a robot that gives backrubs for the low low price of $47 G." Products like this make the Aibo seem like a bargain.
BTW, for anyone interested there is the COG project at MIT, doing research into a humanoid robot, it is VERY impressive.
"Any sufficiently advanced technology is indistinguishable from a rigged demo." --Anon of Ibid (or somebody)
Seriously, though, while I'm not accusing Rodney Brooks & Co. of being frauds, I do think they're getting credit for a lot more than they've really done. Reactive control is wonderful for simple systems. In the MIT robot competitions, the second-place robot is generally some simple reactive design. (At Dartmouth, the first-place robot was, but the last Dartmouth robot competition had a bot construction/program time of about six weeks.)
However, the robot that wins out is generally not using a Cog-like architecture; it's usually one of the designs that uses a more artificial method of intelligence and navigation. The problem with reactive control is that it's easy to build and debug for simple systems, but as you add more functions and layers the connections and interactions become too much for even an MIT mind to handle. Brooks even says as much in one of his early papers on subsumptionism.
There's also Brooks' admission in his own FAQ that even the simple ant robots they've built are utterly unreproducible. IMHO, it's not that great if you can build something once but can't tell other people how to reproduce it.
This isn't meant to be a whole bunch of doomsaying about reactive design --- I think it's appropriate in some circumstances, and I've used it myself when I needed a simple barely-effective hack. It's just that sometimes it feels like the Cog Shop and their academic relations would like people to believe that they're the ultimate solution to all of robotics, and I don't think that's going to happen. It's too hard to use their methods in a general and structured manner to build systems that have anything resembling guaranteed behavior.
Alik
Aging Demes, Wealth and Robots
by
Baldrson
·
· Score: 3
Japan has a problem similar to some demes in the US:
Aging demographics.
At some point, someone is going to have to pay for a lot of domestic servants.
The earliest boomers, like Clinton and cohort, caught the real estate boom of the 1970s. I put the catastrophe point for birth year in the early 1950s, possibly as early as 1950 as speculation from the GI generation anticipated the demand from boomers. Swelling real estate values combined with inflation-depressed real wages in the 1970s and scarcity of entry-level jobs to make being born before 1950 a very shrewd business move (fixed mortgate rates hit 19 percent in 1981 and being a fresh-out of college system level programmer for a major computer company paid around $17,000 in the late 70s.).
The earliest boomers caught this wave, built by their younger siblings, but the largest beneficiaries of all the "eat the boomers" hysteria were the GI generation real estate speculators that liquidated the savings and loan system. The largest wealth transfer in history is now occuring as dying GIs naturally favor their shrewd eldest children among the boomers. After all, they did demonstrated the business savvy, depth of character and wisdom to be born before the 50s.
This means the Clinton/Gore cohort are going to have a big gob of money as they approach retirement.
As they near retirement, these the earliest of boomers will be more and more interested in domestic servants. Whether they'll be able to put up with a mere machine rather than a living, breathing human to dominate is the real question. They did, after all, develop a taste for bossing around huge numbers of boomers who seemed to need it so much (so screwed up they couldn't land a job that could pay for a house and a family -- birth control pills and abortions were far more affordable, the way the GIs and earliest boomers did.
I don't have Sony's market research division at hand.
So, perhaps the Japanese are playing a game of offering strange toys like this "massage/errand boy" robot as a means of feeling out how much of the earliest boomer/GI-legacy gold mine they can grab with machine lackies.
The Japanese need pervasive automation even more than the aging US population because the Japanese are less willing to import labor from other nations. Grabbing the gold with these expensive toys may be the way they finance real automation technology that they desparately need.
Re:Aging Demes, Wealth and Robots
by
gargle
·
· Score: 2
On the topic of machines taking care of the elderly in Japan, there's a very interesting Japanese movie/satire on this issue, "Roujin Z" (with script written by Katsuhiro Otomo of Akira fame).
I've met Brooks, I've read his papers, and I've been out to his lab. I liked the original subsumption/insect level work. But Cog? No.
When Brooks first proposed Cog, his presentation was along the lines of "we don't know how to do human-level AI, so we're going to build a humanoid robot, throw 30 MIT PhD theses at it, and see what happens." I thought this was bogus. I asked him, "Why not go for a lizard or a mouse brain, now that you've done insects"? His reply was "Because I don't want to go down in history as the man who created the world's greatest artificial mouse".
And that's the problem. The goal of Cog is PR, not progress. Hence the Rod Brooks World Tour, Rod Brooks T-shirts, Rod Brooks TV specials, etc.
More fundamentally, a major problem in AI as a field is that people keep trying to make it to human-level AI in one step, rather than clawing their way up the evolutionary ladder with ant-level AI, bee-level AI, lizard-level AI, rodent-level AI, and only then attempting higher mammals. The Big Project to Reach Human Level AI in One Big Step people have gone down in flames decade after decade: McCarthy in the 1970s, Feigenbaum in the 1980s, and Lenat in the 1990s. By now, it's clear that it's a hard problem. Many really smart people have beaten their heads against the wall on this one. Attacking it as if it were one step from solution is doomed.
There's hope, though. Game AI developers are struggling to build characters that can survive in tough environments. They're getting better at it. Work in the game AI field is getting to be better than academic AI; the gamers have a real problem and a real market.
Game AI developers are struggling to build characters that can survive in tough environments. They're getting better at it. Work in the game AI field is getting to be better than academic AI; the gamers have a real problem and a real market.
As someone who has done some work in game development and more work in "academic AI," (I currently work at NASA, applying AI to space missions), I disagree with this statement. The goal of game developers is to write fun games, with a focus on playability. Developing intelligent adversaries has some value, but a truly intelligent adversary would be no fun to play against because it would always win -- it has the advantage of speed and infinite patience, so it needs to be a little weaker in terms of strategy. Add to this the fact that most computer players cheat (e.g. have access to information a human player doesn't) and the fact that many games consist of a single human against a buttload of AIs, and it becomes clear that the AIs can be as dumb as bricks and still make for a great game. I'm not saying that a lot of clever work doesn't go into designing the AIs, just that it's not the perfect problem domain that one would think. Also, from the game developers I've talked to, I don't get the sense that there's much cutting edge AI research going on, though I'm sure there's some I'm not aware of.
According to the pressrelease the robots is mainly controlled via a remote control (or PC).
This demonstartes one of the major problems of robotics to date :
Robot Navigation in umodified environments.
Lots of research goes into the area, but self localisation in an umodified home environment, where lots of obstacles exist, some of which keep moving around (i.e. humans, small furniture pieces, etc) is still a big problem.
The same goes for the use of robotics vision. Yes there are some clever schemes to use stereo cameras(to estimate distance) for object detection and avoidance and some help in landmark recognition, but it takes a lot of processing power and is only mildly reliable. Identifying specific people without major constraints (i.e. person has to face the robot from a certain angle without moving) is AFAIK also not reliably possible.
This are the rasons why at the moment all "proper" (i.e. learning, and not remote controlled or having hardwired complex behaviour) commercial robots are toys. If your AIBO moves in the wrong direction, runs into you, or doesn't manage to fetch its pink ball, nobody will complain it doesn't work properly, people will enjoy it, and find it highly amusing.
Reliable, intelligent, robot servants are still a long way off, but my guess is that they surely will exist in 20 years or so and most likely they won't look anything like humanoid at all...
BTW, for anyone interested there is the COG project at MIT, doing research into a humanoid robot, it is VERY impressive.
Take a look at some of the videos to see what they are already capable of. Especially the head demos are impressive.
http://www.ai.mit.edu/projects/cog/video_index.htm l
People seem to be forgetting that the thing might actually give really damn good backrubs.
$47k, for a lifetime of wicked backrubs sounds okay to me.. Only problem would be tinking with my Robie Jr. until he could handle the thing on its own.
------
-- If a tree falls on an anonymous coward yelling 'first post' in the forest, does anybody hear?
Slashdot Mirror
Here are some links after quick search on the net.
l b iztech/prom/92259
Sorry, only in Japanese. But you should be able
to check out some pics on their site.
Tmsuk Inc. was incorporated by Thames Inc.
Tmsuk Inc: http://www.qbiz.ne.jp/tmsuk/
Thames Inc: http://village.infoweb.ne.jp/~thames/index.htm
Images of robot: http://www.qbiz.ne.jp/tmsuk/image_data/tm4_04.htm
NikkeiBP's article: http://biztech.nikkeibp.co.jp/wcs/leaf?CID=onair/
Additional info on this robot.
size:
length - 750mm
width - 600mm
height - 1,200mm
weight - 100kg
# of joints:
head - 2, hip(?) - 1, arm - 7x2=14, hand - 3x2 = 6
mobility:
two individually controled wheels
front and rear supportive wheels
max speed 3km/h
visual:
250,000 pixel CCD camera, horizantal view angle, 114 degrees, 10 frames/s (w/ current PHS setup)
voice:
4 voice (whatever that means)
sensors:
# of proximety sensor - 5
power source:
Ni-Zn battery (1.5 hours of continuous operation)
Controller:
- joy stick controller for fingers (w/ force feed back) = 4 controls x2
- arm control = 6 controls x2
- head control = 2 controls
- wheel control (pedal type) = 3 controls
- network = PHS (64Kbps PIAFS 2.0)
- display = headmount display
Price: 5,000,000yen
BTW, for anyone interested there is the COG project at MIT, doing research into a humanoid robot, it is VERY impressive.
"Any sufficiently advanced technology is indistinguishable from a rigged demo." --Anon of Ibid (or somebody)
Seriously, though, while I'm not accusing Rodney Brooks & Co. of being frauds, I do think they're getting credit for a lot more than they've really done. Reactive control is wonderful for simple systems. In the MIT robot competitions, the second-place robot is generally some simple reactive design. (At Dartmouth, the first-place robot was, but the last Dartmouth robot competition had a bot construction/program time of about six weeks.)
However, the robot that wins out is generally not using a Cog-like architecture; it's usually one of the designs that uses a more artificial method of intelligence and navigation. The problem with reactive control is that it's easy to build and debug for simple systems, but as you add more functions and layers the connections and interactions become too much for even an MIT mind to handle. Brooks even says as much in one of his early papers on subsumptionism.
There's also Brooks' admission in his own FAQ that even the simple ant robots they've built are utterly unreproducible. IMHO, it's not that great if you can build something once but can't tell other people how to reproduce it.
This isn't meant to be a whole bunch of doomsaying about reactive design --- I think it's appropriate in some circumstances, and I've used it myself when I needed a simple barely-effective hack. It's just that sometimes it feels like the Cog Shop and their academic relations would like people to believe that they're the ultimate solution to all of robotics, and I don't think that's going to happen. It's too hard to use their methods in a general and structured manner to build systems that have anything resembling guaranteed behavior.
Alik
Aging demographics.
At some point, someone is going to have to pay for a lot of domestic servants.
The earliest boomers, like Clinton and cohort, caught the real estate boom of the 1970s. I put the catastrophe point for birth year in the early 1950s, possibly as early as 1950 as speculation from the GI generation anticipated the demand from boomers. Swelling real estate values combined with inflation-depressed real wages in the 1970s and scarcity of entry-level jobs to make being born before 1950 a very shrewd business move (fixed mortgate rates hit 19 percent in 1981 and being a fresh-out of college system level programmer for a major computer company paid around $17,000 in the late 70s.).
The earliest boomers caught this wave, built by their younger siblings, but the largest beneficiaries of all the "eat the boomers" hysteria were the GI generation real estate speculators that liquidated the savings and loan system. The largest wealth transfer in history is now occuring as dying GIs naturally favor their shrewd eldest children among the boomers. After all, they did demonstrated the business savvy, depth of character and wisdom to be born before the 50s.
This means the Clinton/Gore cohort are going to have a big gob of money as they approach retirement.
As they near retirement, these the earliest of boomers will be more and more interested in domestic servants. Whether they'll be able to put up with a mere machine rather than a living, breathing human to dominate is the real question. They did, after all, develop a taste for bossing around huge numbers of boomers who seemed to need it so much (so screwed up they couldn't land a job that could pay for a house and a family -- birth control pills and abortions were far more affordable, the way the GIs and earliest boomers did.
I don't have Sony's market research division at hand.
So, perhaps the Japanese are playing a game of offering strange toys like this "massage/errand boy" robot as a means of feeling out how much of the earliest boomer/GI-legacy gold mine they can grab with machine lackies.
The Japanese need pervasive automation even more than the aging US population because the Japanese are less willing to import labor from other nations. Grabbing the gold with these expensive toys may be the way they finance real automation technology that they desparately need.
Seastead this.
When Brooks first proposed Cog, his presentation was along the lines of "we don't know how to do human-level AI, so we're going to build a humanoid robot, throw 30 MIT PhD theses at it, and see what happens." I thought this was bogus. I asked him, "Why not go for a lizard or a mouse brain, now that you've done insects"? His reply was "Because I don't want to go down in history as the man who created the world's greatest artificial mouse".
And that's the problem. The goal of Cog is PR, not progress. Hence the Rod Brooks World Tour, Rod Brooks T-shirts, Rod Brooks TV specials, etc.
More fundamentally, a major problem in AI as a field is that people keep trying to make it to human-level AI in one step, rather than clawing their way up the evolutionary ladder with ant-level AI, bee-level AI, lizard-level AI, rodent-level AI, and only then attempting higher mammals. The Big Project to Reach Human Level AI in One Big Step people have gone down in flames decade after decade: McCarthy in the 1970s, Feigenbaum in the 1980s, and Lenat in the 1990s. By now, it's clear that it's a hard problem. Many really smart people have beaten their heads against the wall on this one. Attacking it as if it were one step from solution is doomed.
There's hope, though. Game AI developers are struggling to build characters that can survive in tough environments. They're getting better at it. Work in the game AI field is getting to be better than academic AI; the gamers have a real problem and a real market.
It's too bad the article didn't have a picture, so here's a link to one. Just imagine those arms massaging your back....mmmmm.....
According to the pressrelease the robots is mainly controlled via a remote control (or PC).
m l
This demonstartes one of the major problems of robotics to date :
Robot Navigation in umodified environments.
Lots of research goes into the area, but self localisation in an umodified home environment, where lots of obstacles exist, some of which keep moving around (i.e. humans, small furniture pieces, etc) is still a big problem.
The same goes for the use of robotics vision. Yes there are some clever schemes to use stereo cameras(to estimate distance) for object detection and avoidance and some help in landmark recognition, but it takes a lot of processing power and is only mildly reliable.
Identifying specific people without major constraints (i.e. person has to face the robot from a certain angle without moving) is AFAIK also not reliably possible.
This are the rasons why at the moment all "proper" (i.e. learning, and not remote controlled or having hardwired complex behaviour) commercial robots are toys. If your AIBO moves in the wrong direction, runs into you, or doesn't manage to fetch its pink ball, nobody will complain it doesn't work properly, people will enjoy it, and find it highly amusing.
Reliable, intelligent, robot servants are still a long way off, but my guess is that they surely will exist in 20 years or so and most likely they won't look anything like humanoid at all...
BTW, for anyone interested there is the COG project at MIT, doing research into a humanoid robot, it is VERY impressive.
Take a look at some of the videos to see what they are already capable of. Especially the head demos are impressive.
http://www.ai.mit.edu/projects/cog/video_index.ht
Frank
I think you mean 47K. I read "47G" as "47 gig." (hey, this is a geek site). Approx. 5 x 10^10 dollars is a lot of money for a robot!
People seem to be forgetting that the thing might actually give really damn good backrubs.
$47k, for a lifetime of wicked backrubs sounds okay to me.. Only problem would be tinking with my Robie Jr. until he could handle the thing on its own.
------
If a tree falls on an anonymous coward yelling 'first post' in the forest, does anybody hear?