Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
×
Technology

Interview with The Mind Behind Aibo 73

Ant sent us an interesting interview with the man behind the Aibo, Dr. Toshitada Doi. He heads the Digital Creatures Lab at Sony, talks about the history of the Aibo, and where he sees the future of pets going. Speaking as a person who shares his office with an Aibo I think they're neat, but there's still a lot of work to be done before they really break into the mainstream.
This discussion has been archived. No new comments can be posted.

Interview with The Mind Behind Aibo

Comments Filter:
  • Wow. Talk about a short article. I was really hoping for more from that one! Anyone have some really good links to articles, interviews, etc. concerning Aibo, and it's creators?

    Everything I have seen so far has been very interesting - including a nice article concerning how attached to an Aibo someone really can become (the writer had just sent his Aibo back to have it's legs fixed). I want more! (And I want the $2500 to blow on an Aibo, instead of my house or my computer hardware ;-)
  • ..how do you train it to get your newspaper from the newsagents ?

  • I still see no use for an electronic dog, even though their sophistication is simply amazing. I never thought I'd see machines playing soccer for about 50 more years.
  • During a demonstration of the dog here at CNET, we found ourselves involuntarily talking to the robot like we would to a real dog

    Last time I checked, talking to a robot isn't anymore retarded than talking to a pet. It's actually equally fscking retarded.

    Moderate me down ...


    --

  • Well, I put it on my Christmas Wish List.

    It is just too expensive right now. I am curious if anyone on this site will actually get one. I would love to know how they work.

    I wonder how autonomous they really are.

    Right now, I do not have the time or location for a real pet. One that does not eat, shed and shit sounds pretty good to me. Hopefully I could do a better job of keeping an AIBO alive than my plants.

    The idea of raising an abused and unloved AIBO that runs for its life at the site of a pink ball also makes me laugh. Sorry.
  • when they attack the postman delivering the first post.
  • If these eletronic pets become successful, then the same technology, give tweaks, could be adapted to humans also. What are the ethical issues involved? Would YOU want to live the your life as that dude in the Movie multiplicity?

    Please visit FreeDonation.com [freedonation.com] - You can donate Food and Medicine for FREE to Save Children (You can donate once EVERY DAY). The donation is fully paid by corporate sponsors with the money they would have spent anyway on advertising. There is no charge to you.

  • The interview got it's numbers wrong. Only 15,000 have been sold, not 150,000.

  • by A Big Gnu Thrush ( 12795 ) on Tuesday November 30, 1999 @07:24AM (#1493251)
    One important fact the article missed, is that the target market for these pets is not necessarily the US. Pets are far more rare and expensive in Japan. Even on a modest income, I could afford acres of land in (rural) America, and stock that land with cheap dogs. In Japan, this would not be possible.

    I heard a news story on NPR ( sorry, no link ) that talked about a service that let you rent a dog for the afternoon. You didn't get to take it anywhere, you just walked it around the provided area, but you got to pick the dog you wanted.

    Not to say that Americans won't buy this. I would love one, but in Japan, it may actually be a far more practical solution.
  • The dog uses the Aperios real-time operating system. More information is available on the Sony site [sony.co.jp].

    I thought it was particularly interesting that it is IPv6 ready [keio.ac.jp]. Not many people can claim that for their pets! :-)

  • Well, they're not Daggets (sorry Boxy), and Aibo's no K-9 (though Romana can no doubt fix that), but I think it's an interesting concept.

    However, I think they're going completely the wrong way about it, to be honest. The more you pre-program, the more rigid and stupid the machine is going to be.

    There's really no need to hardcode more than the absolute basics. Anything & everything else could (and should) be learned, with the ability for the software to adapt as necessary.

    No, we're not talking Terminator or HAL-9000 technology here. A large, FAST neural net isn't difficult to code. Place the software in virtual reality, for a while (simulate as much time as you like), then dump into the robot's computer. Not that complicated to do.

  • In the context of the ongoing question of how the internet/technology growth foils closeness and communication of people, how does getting programmable robot dogs fit into this? Not enough falseness on the net, so we bring it into our homes?
    But surely that's an overreaction, it's just another hobby, albeit like paying for all of the food your dog will eat in a decade up front...
    I just have to wonder how much wonder that there is present in the real, natural world, we're slowly losing contact in our growing fascatination (and growing market ability to be more so with more product lines) with the technological wonders around us...
    Blah.
  • by cybaea ( 79975 ) <(moc.aeabyc) (ta) (enalla)> on Tuesday November 30, 1999 @07:30AM (#1493255) Homepage Journal
    It is good to see that Sony have finally delivered on what they have been talking about for so long. This year-old techweb [techweb.com] story give a very good background to the project, and discusses some of the more interesting design decisions, including the Aperios Real-Time OS [slashdot.org].

  • or when they core dump on your living room carpet while you are gone.
  • I never thought I'd see machines playing soccer for about 50 more years.

    Really??? Aibo can play soccer for 50 years? WoW! I never thought the batteries would last that long! Are we going to have century-long robotic soccer leagues now? :-)

  • by Croaker ( 10633 ) on Tuesday November 30, 1999 @07:39AM (#1493258)
    This reminds me of Philip K Dick's Do Androids Dream of Electric Sheep. In the future, animals were so rare, that everyone had synthetic pets. Only the rich could afford real pets...

  • Probably human "pets" will be next. As they are harder to build, you'll have to do with Webbie Tookay [illusion2k.com] (get it?) for now. Not the same thing, I know.

  • by r2ravens ( 22773 ) on Tuesday November 30, 1999 @07:40AM (#1493260)
    A friend of mine works in a research lab of an educational institution. They got one of these things and I got to play with it for a while.

    My friend told me that they had spent considerable time training the thing to walk and to react by praising (pressing and holding the large button on it's head) it when it did something right or desirable and punishing (tapping the same button) it when it did something undesired.

    It walks and if it falls over (or is pushed ;) it stands right back up. This one will put one paw in the air as if waving, it stretches and sits and lies down. I will react to people and things in it's environment, although we couldn't get it to play with it's little pink ball. Of course, as the FAQ's for the dog say... "Q: AIBO won't play with his pink ball. A: Maybe AIBO doesn't *want* to play with his pink ball."

    It appears to have some level of intelligence and does appear to learn as it goes. If money were no object, I would love to have one of these. It's fun. When you don't want to play with it, you don't have to feel guilty. And it's a wonderful chick magnet. ;)

    The next version will be even cooler as when it's battery begins to run down, it will seek and go to it's charging base to recharge itself.

    It's an interesting toy for now, but too expensive. I do agree that 10 to 20 years from now, this will be very common in peoples homes.

    And just think, Asimov was right again...

    Russ
  • So what would one cost then? I want one.. err.. I NEED one ;>
  • Arsenal nil, Aibo 2 after 25 years of play. Now for the half time wrap-up :)

    Pope
  • by penguinicide ( 73759 ) on Tuesday November 30, 1999 @07:45AM (#1493263)
    Here are a few links that you might find interesting. Xrays of the dog. http://www.nnc.ne.jp/~as212/aibo/x-p.html [nnc.ne.jp] Aibo disassembled. http://www.aibosite.com/index-s.html [aibosite.com] Aibo taken apart. (in japanese) http://web.kyoto-inet.or.j p/people/nktks/TEMP1/menu1.htm [kyoto-inet.or.jp] Robotic Cat - (evolved neural net) http://www.genobyte.com/robokoneko.html [genobyte.com]

  • Try this [genobyte.com] on for size.

    Exactly what you ask for, but in a cat [genobyte.com].

    Rumor has it they plan on turing it into a consumer robot sometime in the next year or so.

  • by dieMSdie ( 24109 ) on Tuesday November 30, 1999 @07:57AM (#1493265)

    CmdrTaco and/or Hemos could write an essay about life with an AIBO. I'm sure that between the two of you, there are plenty of amusing stories accumulated already...
  • After the (apparent) commercial success of the Furby and the AIBO, what do you see as the future of commercial autonomous robotics?

    The Kulturwehrmacht [onelist.com]
  • If this is the man behind the AIBO, then what is he doing back there?

    Hey - leave that thing alone, GET OFF!

    Hotnutz.com [hotnutz.com]
  • You bring up an interesting idea:

    No, we're not talking Terminator or HAL-9000 technology here. A large, FAST neural net isn't difficult to code. Place the software in virtual reality, for a while (simulate as much time as you like), then dump into the robot's computer. Not that complicated to do.

    While the flexibility of neural networks is great, they need to be trained (Dead on, on the offline training btw). The bad thing: When training that neural network you need to be working towards a goal.

    How do you define that goal? You've got (assuming some sort of feedforward or derivative network) an output layer generating some output vector, and you're comparing it to an established goal vector. How do you define a goal vector for a behaviour? Is the euclidean distance from your output vector to the goal vector a valid metric for determining error? I can see that being very unpleasant to formulate...

    Not that it wouldn't be a cool research project...

  • by A Big Gnu Thrush ( 12795 ) on Tuesday November 30, 1999 @08:14AM (#1493271)
    Starting to drift off topic here, but you mentioned one of my favorite books...

    One added element in DADOES, it was considered bad citizenship not to own a live animal. It was everyone's duty to own and care for animals, but average people could not afford anything larger than small pets. To prove their citizenship and empathy (read: humanity) people displayed fake animals on the roofs of their houses.

    This book is so good it makes me want to cry.
  • by Anonymous Coward
    I have watched the whole AIBO thing with interest,
    especially when comparing real dogs to AIBO.

    Everyone forgot one thing...a real dog shows geniuine affection for its owner (assuming your not a cruel abusing bastard). Some pet owners, myself included, would say our dogs love us, and we love them, just like any other family member.

    AIBO is a neat toy, but when the day comes that robotic dogs, or people for that matter, have feelings, then what do we do? Or if their simulated intelligence begins to challenge our own?

    "You want me to chase this stupid little pink ball? I dont think so. How bout you go get the oil can and the battery charger for me, or I'll break you neck".

    Heh.

    "I know what love is Jenny"
  • Does anyone wonder if he looks like Mr. Roboto from that STYX video? (Apologies to anyone under 30 who doesn't get it) ;)

  • That would be perfect. You'd have to give him that antenae thing he has though, and the laser weapon. The PBS station here in Albuquerque is showing the episodes with the 1st Romana right now and I'm having all kinds of good dreams about her and the privacy of the Tardis. :)

    I know this will get moderated down but my karma can take it and I needed my geek fix for the day.
  • I keep thinking of the guys behind [cyberlife.co.uk] the incredibly cool Creatures [creatures3.com] game series. There is a fantastic article on "cyberlife" [creaturelabs.com] on their site by Stephen Grand that's surely worth reading if you're into AI/alife.

    I'd like to see the Cyberlife people team up with the Sony people and put their seriously advance AI code into a hardware device. (Wow, how about a "real live" Norn? That would be awesome. They'd just have to have a longer lifespan than that in the game - about 8 hours. :)

    I think the most interesting aspect of the Cyberlife technology is the synthesis of a "biophysical" system in software - they have code to simulate digestive, circulatory, immunological systems, etc, completely outside of the neural network that makes up their cognitive system.

    If you could marry that concept with hardware to emulate it (i.e. the battery is getting low so the feedback to the brain is "I'm hungry!") I bet you could come up with some seriously complicated and complex emergent behaviours.

    -=-=-=-=-

  • I sure hope that Mr. Moderator wasn't referring to his .sig as flamebait... that would just be opening up a whole other can of worms.
  • You will have to wait a year or so, or purchase on on an online auction. Ebay and Yahoo will have them. They seem to be going for around $4000

  • Judging by some of the "Anonymous Coward" posts I've seen lately, it's already happened.
  • There is a website, created by an Aibo owner, that is a meeting point for all Aibo owners.
    They show off their photos, discuss various problems, remark about the AI, etc..
    The website is called Aibonet [aibonet.com].

    The Aibo is THE girl magnet for geeks. I've seen a crowd of women cooing over an Aibo that some guy owned.
    Aibo does fetch...just not sticks ;)

  • Well, they're not available at all, but if you have $2,500 it looks like eBay prices have settled down a bit, you might be able to get one for around that, or for a very slight premium.

    D

    ----
  • An interview with the man inside the AIBO?
    I don't buy it. Admittedly I believed the
    little people inside the TV thing until I was
    twelve, but I caught on eventually. I'm sure
    there's just computers and stuff in there.

    ~yair

    Oops. Never mind. Reading is hard.
  • Actually, the current version has "recharge" as one of its instincts. Although it cannot get onto its charging base by itself, it will ask you to when its batteries get low.

    Aibo works by 4 different instincts, and 6 different emotions. The emotions are (I believe) percentages - so Aibo could be at any given time 10% angry, 20% loving, and 70% curious.

    More information about aibo can be found at this site [aibonet.com], which includes a link to Sony's official Aibo site.

  • What this dog really needs is some nice fake dog fur instead of a silver plastic body. Believe me, it will make a big difference to kids. They'll like playing with the "doggie" instead of the "robot doggie."
  • You only need a stated goal if you're training the net directly. There is a way, though I forget the details, to train certain classes of neural net without requiring a preset list of outputs.

    However, let's assume you do use an absolute list. This -does- get complex, as your output is a time-dependent matrix.

    In other words, your error is sampled not just over an instantaneous output, but over a period of time.

    Let's take an example. Say your robot dog is trying to walk from one end of a room to another. It encounters an obsticle. It's no use the dog just turning left. Yes, it's no longer obstructed, but it's now going in the wrong direction. BUT, it's ONLY the wrong direction because it didn't turn back on course, once the obstruction was avoided.

    An equally valid solution would be to turn right, and walk around the obstruction that way. Or to jump over, if the dog is capable of that.

    Once the robot has made it's manoevers, the external error is in the angle away from reaching the final objective. This needs to be mapped onto the matrix for the entire sequence, so that you can back-propogate, to correct the neural net.

  • Maybe in future models, it will... plug into an IP port, download posts to your favorite newsgroups, then follow you around reciting the postings using text-to-speech.

    [humor]
    The way how some of the /. posts can make me so angry... I'd only end up kicking the crap out of that poor aibo receiting posts.
    [/humor]
  • Certainly I interprete affection from my pet cat's
    actions, but I don't really know what's going
    on inside his head. I don't know how he's
    thinking, or if anything even remotely like
    emotions are happening inside the complex flesh
    computer that's his brain. It's not like it really
    matters to me that I don't understand my cat --
    I often interpret some *very* human motivations to
    my cat that almost certainly arn't really there,
    and it doesn't really bother me.

    That being said, I imagine one of the problems
    with AIBO would be that it's not warm. One of the
    things that makes it hard to snuggle with my
    Iguanas is that they're cold-blooded, and so
    very often they're not pleasant to snuggle with
    because all the warmth goes from me to them. My
    cat is a lot more pleasant to snuggle with because
    he's warm on his own :) I wonder what kind of
    heat tolerances could be designed into an AIBO to
    allow it to have a preferred warmer than room
    temperature body..
  • Look at my other post in this series. There is a project to evolve 32,000 neural net modules containing up to about 1000 neurons each in an electronic brain of a robotic cat. The early stages are emulated in software, including the motions of the cat, gravity, limits of the motion. Later the evolved nets are transferred to FPGA's and placed in a physical robot to contiue the evolution. Here is the link again. LINK [genobyte.com]

  • Most dogs develop and understanding of several hundred words as they mature. Not Shakespeare by any means, but quite an accomplishment anyways.

    I am sure the "first post" couldn't boast such a grasp of the English (or any other) language.




  • Well, I was bit more interested in the (implied) use of neural networks to simulate behaviour (this is also going to be a bit of a reply to penguincide's article link to the Robokoneko :).

    How exactly would you determine when to be, say, "frisky"? Or rather, we have a neural network with a bunch of sensor inputs, some other state data, etc. How do you take that input, run it through a network, and decide that the emotion you'd like Aibo (or Robokoneko) to exhibit is "frisky" --> where you've gone and developed some subsystem (perhaps a neural network) to provide the physical "friskiness" of the robot?

    I took a very quick look around http://www.hip.atr.co.jp/~degaris/papers/ and didn't find anything that indicated how they planned to train the network(s) (Although there was quite a bit of interesting stuff to read). One of the bits of correspondence I found questioned how higher level control (not just basic positioning/simple actions) was going to be done, and I didn't really see an answer to it. Again, I just took a quick look around, I don't have the time (today) to read through all the papers...

    You wrote :

    However, let's assume you do use an absolute list. This -does- get complex, as your output is a time-dependent matrix.

    In other words, your error is sampled not just over an instantaneous output, but over a period of time.

    Let's take an example. Say your robot dog is trying to walk from one end of a room to another. It encounters an obsticle. It's no use the dog just turning left. Yes, it's no longer obstructed, but it's now going in the wrong direction. BUT, it's ONLY the wrong direction because it didn't turn back on course, once the obstruction was avoided.

    In this case, a shortest path algorithm would be much better :) Your net's outputs are driving the velocity/acceleration of motors, so, at any given moment you'd like : the acceleration to be smooth (good to backpropagate errors immediately), and the dog to be heading along a predetermined path (After all, this is what offline training's all about : predetermined data). Hum, this bit's come out all wrong! Anyway, the size/space requirements for most of the input data is pretty small (probably under a gigabyte or so), so offline training is probably viable.

  • 'cos you've obviously learned nothing from the other animals we share this planet with. Get back in ur box and keep ur beak shut
  • There are 15,000 different sets of the AIBO AI being taught how to interact with a human environment. One of the neatest things I've come across reading about how people using their AIBOs was from aibosite.com [aibosite.com]'s FAQ. Apparently some AIBO users have found their AIBOs have developed rudimentary face-recognition, even though Sony claims that no such software was installed in untrained AIBOs.

    This might light up some people's Big Brother radars, but what if those 15,000 trained AIBOs downloaded the product of their training into a central database at which someone(probably Sony, but what are the possibilities of GPLing the training you've given your AIBO?) can sort through the best of the acquired programs and redistribute them to new AIBO purchasers? With so many individuals constantly training(ie, improving the software of) their AIBOs, wouldn't the software increase dramatically?

  • The translation of "aibo" (romaji Japanese), according to Jeffrey's J/E Dictionary [mv.com] defines what qualities we all admire in a real dog:
    -cherish the memory of; yearn for
    -love; attachment; adoration
    -companion
    -partner; pal; accomplice
    Now, this may seem OT to some of you, but I read through various postings here, saying all the Aibo lacks dramatically is the ability to love it's owner back, like a real dog. I just happened to look up the name...and seeing these definitions in correspondance with what was said here on /. it really makes you think : ) At least, it should...
    Mmm...Sony.

    And yes, an Aibo is a great chick magnet. We are known for our expensive tastes ^-^

    miyax

  • Give it a PIII for a brain, and you have an instant heater :)

  • The framework for cybernetic human companions is already in place -- if you combine Aibo's bran and a RealDoll (with a few modifications), I am sure you can get some pretty interesting results. However, a 'human' companion may not be as nice as a dog (or cat) because they would be much larger, and cost would reflect that. I am not sure how much realdolls cost, but you'd have to implement a moving skeleton, and a rather fast computer inside the rib cage, maybe a gyroscope and some cameras and microphones in the head. (the power plug can either be something in the head covered by hair, or if you don't want real genitalia on your pet human robot, then maybe in there somewhere) Anyway, this type of Terminator-ish skeleton, even if spared the silicon boobs, would get expensive beyond belief, so unless you are willing to trade your Dodge Viper for a human that is clumsy, unable to hold up conversation, and good for nothing but "give me that glass" and "Bend over"... well, let's just say, Sony won't be making these :)
  • I can just imagine it barking out spam! ;-)
    --
    Leonid S. Knyshov
    Network Administrator
  • "queer for gears..."
    -- Naked Came the Robot, Barry Longyear
  • Hey, I'm 26 and I got this... Maybe you should lower the age limit a bit huh... :)
  • I'm sure you are a troll but. http://slashdot.org/code.shtml [slashdot.org]

    Vermifax

To be is to program.

Working...