Ok, someone spent a lot of time programming various semi-autonomous moves into a cute little mechanical man, but really, who cares? Is it fundamentally any different from what the animators do every day at Pixar? Excellent entertainment and probably great training for making additional little semi-autonomous mechanical men who can do still more closed-ended stunts. Yawn.
I'll be impressed when a robot, even a very simple one, can fend for itself in the real world by sensing its environment and adapting (like Stanley in the DARPA Grand Ghallenge). The main thing these little "humanoid" robots show is how much time and money people are willing to spend on fancy toys. They might as well be humanoid pencil-sharpeners.
As someone who ran one of the DARPA Grand Challenge teams (Team Overbot), I'm impressed with these little humanoid robots. We're seeing about 70% of the functionality of the Asimo hardware at about 1% of the cost.
The hobbyist humanoid robots are rapidly improving. The better ones now have servos with force feedback info, some rate gyros, contact switches on hands and feet, a camera, and a data link. That's almost good enough to get serious about autonomy. The force feedback info needs to get better, the force sensing needs to be better, the gyros need to become a full 6DOF inertial unit, the vision system needs go stereo, and the data link needs to get up to WiFi levels. Then you have the hardware needed for serious work. Right now, the little humanoids don't have enough data coming back to become intelligently autonomous. That can be fixed, and it's not going to be expensive.
The software still tends to be rather simple, but once such hardware can be purchased for $1000 or so, I expect the software to improve rapidly.
Robot hobbyists, though, are about to face the same transition game designers hit when games went 3D. Suddenly, you needed to know much, much more math and theory to make it work. The game community was able to cope. A few really bright people ground through the theory problems, papers, how-to articles and theory papers were published, code libraries were written, programmers got up to speed on matrix algebra, and soon everybody was doing it.
Then the game community had to do it again when games got physics. That was hard to digest, but that's standard now, too. Game programmers had to learn differential equations, but we got through that.
That's about to happen to hobbyist robotics. You're going to need more math and theory in a year or two to make your humanoid autonomous. Dynamics. Control theory. Adaptive control theory. Vision processing. 3D modelling from data. Simultaneous localization and mapping. All that academic work over the last twenty years is about to be used.
It's going to be really great. Tough, but really great.
What's going to make it really great is that you can afford to make mistakes with these little humanoids. The Asimo-sized machines are so expensive that they can't really do much experimental locomotion work, because the things will fall down. The hobbyists can. Even if you break them, you can usually fix them without spending too much. That's a very good thing.
All right, I'm sorry, that was out of line. I should get a newsreader with a 5-minute delayed posting queue so I don't actually send the first heated response that comes to mind.
Let me try to express that sentiment in a more reasonable manner.
I'm surprised, and somewhat disappointed, by the disparaging remarks a few people here make about even the best of the Robo-One robots (the one mentioned at the start of this thread is a good example). Not too long ago you could count the number of walking bipeds on one hand, and each of them was a research project into which tens of thousands of dollars (if not more) had been poured. In only a few years, these hobbyists have replicated many of the feats of robots that cost orders of magnitude more, and every year they've been getting better.
If one of these disparaging posters had long ago built a bot that could walk on two legs (actively maintaining balance), climb up and down and down stairs, pass a ball from one hand to another, get to its feet from any position in which it might happen to fall (or be knocked), climb across a horizontal pole, traverse a rotating platform, and so on, THEN I'd say, "OK, from his point of view I can see how this is not that impressive." Especially if said poster had managed to do all that for under $10K, as I think most of the Robo-One competitors have done.
But I strongly suspect that, in most cases, the disparaging remarks are NOT coming from these feats being "old hat" at all. Rather, I suspect that what we have here is a bit of jealousy, because some guy who runs a noodle shop in Japan has built a mechanical marvel far beyond what the poster has ever managed. It's the old defense mechanism of trying to build yourself up by tearing the other guy down. It's just not a very effective defense, since it's so transparent.
More charitably, I suppose, it's possible that some such posters are simply interested only in AI, and not in mechanics or dynamics at all. It's true that the AI these robots display is fairly rudimentary, though their mechanics are both elegant and effective. But this is comp.robotics.misc, not comp.ai. If one has interest only in AI, and has no regard for mechanical wonders like these, then I don't understand why one would be reading this newsgroup in the first place.
More practically, I predict (by exptrapolating from observed trends) that in a few more years, Robo-One style competitions will be as popular here as they are now in Japan, easily eclipsing all other types of robot competitions combined. A large number of new hobbyists will enter the field, many of them with no previous robot experience (like Onimaru2's builder). This is the style of robotics that will inspire kids to study science & engineering, and eventually, will make its way into our homes in the form of large humanoid robots that are actually useful.
This is not to say that other types of robotics aren't exciting or useful -- they certainly are (I'm eagerly awaiting the day when my car can drive itself, for example). But attempting to resist what may later be seen as the "humanoid robotics revolution" seems a bit childish. Better to embrace it, and think about how our own particular areas of interest can contribute to, or benefit from, the growing enthusiasm of the public for these amazing machines.
And please accept my apologies for my earlier remark.
I don't deny that the little robots are impressive feats of engineering and artistry that only a rare few have the talent and intelligence to build, but IMO they have very little to do with AI, natural behaviour, or even industrial robots. They are toys in the same sense that styrene plastic car models were toys a generation ago - they require highly developed skills that have little to do with the things modelled. The results are intriguing, but fragile, unstable, and ultimately useless. Little robots may help to teach perseverence, a sense of style, and all sorts of things that enrich the builder's lives, but the actual robots are just expensive throwaways.
Here are some questions that come to mind:
- Why humanoid? To me this is smoke, mirrors and hype. No device that small could ever do the things a real human can do, except as a caricature. They are cute, fun, and impressive but why not a cat-oid or a rat-oid? Not cute enough? Natural behaviour too boring? Try making a robot that can behave like a real rat. No amount of what passes for "humanoid" programming will get you even close.
- What's the attraction of karate, fighting and various other childish obsessions? The resemblance to the offerings at Toys-R-Us is telling.
- What evidence is there that this stuff will inspire future engineers and scientists? It seems more likely to inspire future wrestlers. I have no data, but I suspect that most successful scientists spent more time on school work than on expensive time-consuming hobbies.
Fundamentally completely different. One is real. The other exists only as data. The one in the bit-plane needn't bother with cost, sensors, battery drain, motor stretch, mechanical fatique, or even physics.
As the other Joe pointed out, this is all about inspiration. Woody in Toy Story doesn't inspire the mechanical challenge that even a simple servo-based biped can.
I can agree with that. But then industrial robots don't have much to do with AI, nor does natural behavior have much to do with industrial robots. There are a variety of unrelated subfields of robotics, and no one robot project is likely to address all of them.
This is unfair, I think. These robots are not fragile; the ones that are don't make it very far in the competition. Many "real" research robots don't take very well to even falling down on their own; these ones are actively knocked down and beat upon, repeatedly, and they keep getting back up. Nor are they unstable; again, if they were, they wouldn't be able to take multiple hits without falling over, as the best of these can do. The good ones employ acceleration sensors and automatically adjust their footing and posture to maintain balance -- not a trivial problem in itself. I rather suspect that if ASIMO were subjected to unpredictable external forces, in scale proportion to what these Robo-One units handle, it wouldn't fare as well.
As for useless -- well, let's be honest, almost all our robotics projects are useless in terms of immediate practical benefit. We do them to grow our skills, to inspire others, and maybe someday to contribute to something which can someday be of use. These robots do all those same things too, and more effectively than the little two-wheel-and-caster robot I've managed to half-build myself. Perhaps your own project has more practical value, but that's no reason to disparage those of us who do it primarily for love of robotics.
The Japanese in general feel that if robots are going to be generally useful machines in our world, they have to be able to navigate environments and operate devices designed for humans. That implies legs and hands. I think there is some merit to this view.
Of course not, but that misses the point. Being small keeps the costs down, providing a platform on which far more people can develop new techniques and approaches. The basic problems are the same; being small does not make it any easier to maintain balance, climb stairs, or work a doorknob. So instead of one million-dollar project trying to solve these problems, we have thousands of thousand-dollar projects trying to solve them. That seems like a good route to progress to me.
No, too far removed from the ultimate goal. Cats and rats can do the laundry, fetch the mail, load the dishwasher, rake the leaves, and put books back where they belong on the shelf. A robot to do any ONE of these tasks could have a specialized form, but if you want one household robot that does them all, the human shape is probably best.
Not that these robots are capable of most of that yet -- but if you look at how much progress has been made in the last four years, and compare it to the previous 10 or 20, it seems clear to me that this approach is a productive one.
And conversely, no amount of "ratoid" programming will get you close to a robot that behaves like a real person. So?
Well, I'll admit that the boxing competitions are pretty obvious crowd-pleasers, though they do raise a number of interesting problems (like applying force to another object, and accepting unpredictable impulses yourself, without falling over). But look past those at the other competitions -- stair-climbing and door-opening and so on. And note that they don't build a bot specific to any one task; the goal is always to have one bot that takes the medal in all the various tasks.
It's exactly that sort of generality which is the real heart of robotics, I think. It's relatively easy to build a task-specific machine; just look at the dishwasher for example. What separates robots apart from mere appliances is that they can do a wide variety of tasks, even ones not anticipated when it was built.
The reason people spend so much time on humanoid form, function and the movements based on karate etc is money. The humanoid movement captures the attention of the young and older generation. Attention equals money and money equals funding at schools. More competition equals more attention and more money.
About fifteen years ago, when Rod Brooks had done his insect robots at MIT, he came out to Stanford to talk about "Cog", that sort-of humanoid torso with facial expressions. He was talking this up as a path to human-level intelligence. I asked him, "You've done insect level AI. Why not try for mouse-level AI next. You might be able to achieve that". He replied "Because I don't want to go down in history as the man who developed the world's greatest robot mouse".
Well, Cog was a dud, it didn't produce any significant advance towards human-level AI, but it got lots of PR for a while.
Thanks for writing such an interesting reply, even though it was a month ago. One of my friends is into Robo-One, and wanted to buy a RoboNova as soon as he got some money together, so I've done some research into these robots myself.
I have a long-term goal of building a much more complex robot (based on a velociraptor, using series elastic actuators). However, it is a multi-year project, and it would be neat to have something to play with in the meantime. I looked around, and found an extremely interesting new kit available:
This is an amazing kit, mainly because the servos provided are so interesting and powerful. They have more than twice the torque as the RoboNova stock servos, and they are controlled via a daisy-chained serial system. They have position, torque, and temperature feedback, and can be run either as normal servos or in continuous rotation mode (without hardware changes).
So what I'm thinking, is you get one of these kits, and a gumstix (which I already have) to replace the controller it comes with. Since its a simple serial protocol (and its fully documented in the servo manual), controlling
18 or 20 servos from a gumstix would be no problem. You can run some serious software on a gumstix, plus have a bluetooth (or wifi if you have that model gumstix) socket interface for real-time visualization/feedback/control on a PC.
With the force/position feedback servos, you could do some fairly sophisticated dynamic gait work. Add these parts:
?products_id=637 and a gumstix and you could have a pretty wicked small humanoid for under $2K.
The future you described may be closer than you thought...
-------------------------------------------------------------- Jon Hylands email@example.com
The Bioloid is an excellent kit, especially if you want to experiment with quite a few different configurations and designs. The Bioloid fanatics over on the RoboSavvy forum have been going at it like gang-busters including doing considerable reverse engineering:
They even have one of the support people from Robotis actively participating in the forum.