Slashdot Mirror
New Walking Robot From Honda
Jimbob2001 writes: "Forget the Sony AIBO: this new Walking Honda should be the new must have plaything (sorry, not available for sale!).
Here's the slick PR Web site (with movies etc.), and here's the developers' site with a little more detail. (Note the P3 is the latest version of this long-standing development project.)"
The BBC recently carried this story about polymer muscles for microbots. (Not useful on the macro scale, but still interesting.)
Corollary to Moore's Law: The IQ of new computer owners is declining.
It is a grand public relations stunt. You are correct that this robot is useless to a consumer, but stunts like this are what fund real robotics.
Some of the technology in this robot:
-binocular vision
-miniturized servo joints
-miniturized controller. (Contollers as large as refrigerators are still used in industry.)
-ballence system adjusting to payload
These technologies are not revolutionary, but important evolutions for industrial robots.
Tell your investors the phrase, "evolutionary reducion in controller size". Boring. Tell them, "walking robot" instead. That's news!
Remember, many manufacturers also sell industrial robots. (Kawasaki and mitsubishi are other examples.) We can say, "anthropomorphic robots are useless," but Honda is going to take the technology they developed for it and apply it immeadiately to industrial robots.
Honda has the advantage of being an industrial company with a good mechanical engineering R&D operation. To build something like that, you need more technicians and machinists than researchers. Most robotics labs in the US are in computer science departments, and with the exception of the Field Robotics Center at CMU, aren't organized to build good machinery on a reasonable schedule. DoD isn't throwing money at this problem any more; they did in the 1970s and 1980s, and didn't see much for their money.
The hardware component state of the art is actually pretty good, which wasn't true a decade ago. Early robotics researchers wasted too much time on building radio links, motor controllers, encoders, and similar parts. Now you can buy all that stuff. Getting enough compute power onboard is now the easy part. Rate gyros and accelerometers are now stock, low-cost items. CCD TV cameras are easily available. Laser rangefinders are still big, clunky, and overpriced, but depth from stereo vision, after thirty years of work, now works well in real time.
Controlling a legged robot is a tough problem, but there's been a fair amount of work on balance. I have a patent in that area myself. Most of the work on legged locomotion is now going into animation and games, but the results will be useful in the real world.
Nobody makes money from mobile robots, though. A few companies have tried, notably Denning Robotics and HelpMate, but not with success. The basic problem is that robots compete with cheap people, and aren't much faster.
Nitinol is made from Nickel and Titanium, and changes its shape based on applied current (well, "Ohmic Heating", actually). A search on Google turned up such cute examples as this.
It doesn't seem strong enough for large-scale use, but it does show proof-of-concept nicely.
To think about it, the robot imitates what humans do well - it can walk, but it is not necessary that robot has to imitate everything that humans do and rely on simple video image to imitate vision. Robot can be smarter than humans, it can use resources unavailable to humans that are way more precise: lasers, sonic, microwaves, metal detectors, night vision etc. A robot that uses more than one input device to figure out the surroundings exactly can be so much more useful than a fully blown AI system that completely "IMMITATES" what humans do - rely on texture processing to figure out surface shape and depth. A robot that relies on textures can not be as efficient as the robot that actually "KNOWS" the shapes and depths and even compositions and various physical measurements such as strength of materials. Sending the latter robot to Mars will be more productive than sending the former human immitator.
You can't handle the truth.
I see 3M as a potential contender, but I have no idea how far any research has progressed into this field.
Once somebody makes such a material, it would only be a matter of time before hobby robotics really take off!
I can see it now, people going out there to buy metal rods, that are held together in a ball-socket joint configuration, and then the muscle strands attached to both pieces of metal, and then wired to the microcontroller. And ofcourse the OS would have to be open-source, Robotix, or just Linux for Robots... Ofcourse Microsoft may still be around, and release Windows RE.
You ask why it's not easy to find stairs.
Among other things, consider the problem with disparity: Looking down gives you horizontal disparity, i.e. the image from one eye is shifted slightly from the image of the other.
Problem is, stairs are horizontally aligned, so there's no "bright line edges" to detect distance from.
Instead, you have to do what the brain does, and search for texture disparity. Good luck; we don't even have compression algorithms that approximate high frequency textures(Perlin Noise isn't hugely flexible nor reversible from real life textures, though I'd wager it could be). To do good binocular disparity on a texture, we need the ability to say, "If the texture was 5 feet away, the two surfaces would differ by x. If the texture was 10 feet away, the two surfaces would differ by y. Now, lets compare these two intrinisically noisy images across multiple texture sizes and detect where in between 5 and 10 feet we are."
Actually, that shouldn't be impossibly different, but it's a hell of alot harder than poking around for a ping pong ball.
Yours Truly,
Dan Kaminsky
DoxPara Research
http://www.doxpara.com
Genetic programming is much more promising than programmed control, but I wouldn't call it learning. It's a means of finding an algorithm that works without writing it yourself.
If genetic programming was learning, you should also be able to say things like "Mankind has learned to have two arms and two legs.", and to me that sounds wrong... more correct to say that Elvis has evolved to walk.
Oh great, a fully mobile humanoid robot. Now they just need to stick Windows on there and let it go around killing people and terrorizing the city.
//m
Besides, there is even a solution to the problem that doesn't involve an extra camera. You could simply look at the frames from a few seconds ago, and compute the disparity using them. The old frames will be displaced along the other dimension if the robot has moved forward.
I wonder what a person would do faced with a perfectly horizontally symmetrical scene...move their head and eyes around in an attempt to establish binocular disparity?
--Dan
Finally I have someone that can walk my Aibo in my absence! :)
----------------------------
----------------------------
Esobofh - Currently drinking fresh mango juice.
Lots of people said the same thing about going to space and the moon. But look at all of the spinoffs that came from NASA's efforts, like teflon and stronger ceramics and plastics, etc... The developments along the way are just as important as the finished project.
>P3's carbon rucksack also holds the battery.
I coulda swore the first time I glanced at it, it said "carbon nutsack"
Need Free Juniper/NetScreen Support? JuniperForum
Where did they find someone skinny enough to fit into the robot suit?
turn up the jukebox and tell me a lie
You can find a Swedish robot project at http://humanoid.fy.chalmers.se/.
Featuring Elvis (named so because he moved his hips when he learned to walk), who is a tad bit smaller but has learned to walk all by himself (genetic programming).
The Honda walking robot is one example of what many scientests feel is the way forward with regard to artificial intelligence and cybernetic systems development: Instead of trying to replicate core intelligence, many feel that the way to go is to replicate physical functions - IE, to create a "synthetic humanoid", instead of trying to create a "Brain Box", or neural network emulating human synapses and neurons, rather create a "human frame" that duplicates human movement, and from there, slowly "teach" the "organism" using various methods. Even if it comes out as a very slow, sub-ape intelligence-wise, some feel it will be making more progress than neural network/software brain modelling.
I'm of the opinion that the brain is a system like any other, and, like any other system, will eventually have documented interfaces that scientests will be able to write against, creating an emulator, and eventually, a compatible interface. A lot of research going on at my current firm deals with software interfaces for biological structures, as well as analysis of the said biological structures. The future will, have no doubt, be very interesting."A few atoms won't even light a match" - Dr Jones, 1933
http://www.cs.cmu.edu/~robosoccer/
One of the major problems with this robot (I've seen it before, several times over the last few years) is that it requires a massive amount of power. You can only fit so much juice on the machine before the container of said power itself becomes too much of a physical burden on the robot to reasonably pack.
---
seumas.com
It isn't sexy and is probably cheating for those who demand strict anthropomorphism, but as a cash-strapped teen hobbyist in the 70's I found that edge detection could be done easily with a cheap IR laser diode casting a stripe at a 45 degree angle (you could probably use an IR LED, but I needed a *bright* dot, or my crappy vidicon tube setup couldn't distinguish in lighting like 'sun streaming through venetian blinds')
It didn't scan and it wasn't pivotable. (Both were planned future upgades that never got done) It was just a fixed optical element that converted tha IR laser 'dot' to a 'stripe'. It could rotate, so that the stripe could assume different angles to the vertical, but +45 and -45 were almost always adequate. The one trick is that the 'stripe' was a better discriminant when projected from near or below the level of the steps, not at head height. The video unit could be anywhere.
IR showed up nicely on the old tubes I used, but was chosen because that's all I could get ($10 a pop *surplus* for the diode alone in 1978). It did made the system look cooler to human eyes, though. A trio of different colored visible light laser stripes would've been a very distinguishable signal in high noise, but that was just a dream back then. Now you can buy color laser pointers with sets of removable holographic gratings for a few bucks. I bet a simple fixed grid holograph at 45 degrees would do the job nicely. A second at 22 degrees would be a great backup.
Subtracting successive beam off/on frames gave me all the info I needed for edge detection with monocular vision. Binocular should give you everything you need to climb stairs, I'd imagine.
Admittedly, the discontinuity detection was more processor intensive than an edge filter, but I'm sure there are more efficient algorithms than the ones I used (and there simply is no comparison between an 0.5MHz 6502 and a GHz Athlon)
The question is: would you rather be totally anthropomorphic or just get the job done?
This approach probably wouldn't work for industrial robotic assembly (which may be why Honda didn't use it). Shiny surfaces, like factory fresh metal parts, really kill the image (and 'beams bouncing everywhere' wouldn't be too kind to bystanders, unless you stuck to low power IR)
If you can go to bed, knowing you did a valuable thing today, you're very lucky. If you can't... it's not bedtime
If you are *nix user, don't bother downloading the movies. They are Sorenson all the way. Pity, such a high-tech site denied to high-tech people....
www.eFax.com are spammers
it needs binocular vision to fit its "human environment" design brief, but why can't its hearing be tweaked to make up for the defficiency? Humans who have limited sight may compensate with other senses, so it would still be ok conceptually. but presumably untrasonic sonar could give a complete 3d image of the room, and possible info about surfaces, densities, maybe even the velocities of moving objects..
--
So Honda's got a walking robot (Honda Germany, I'll note) and Volkswagen has a car-driving robot. How long before we see robots driving taxis a la Total Recall?
And, just to throw in some eco-geekiness, it'll be a fuel cell-powered taxi, to boot.
Well, for honda, anthropomorphic is obviously the goal.
1. Anthro has more promo value. A cart that just wheels around looks too much like the Radio Shack remote-controlled toys to John Q. TVwatcher... a robot that walks and acts like a human, otoh, gets a big wow for honda. This is imporant, since bleeding-edge research projects get canned unless they can a) promise sell-able, profit-making results in the forseeable future or b) generate enough jaw-dropping to be considered good PR. In the post-reganomics world, projects that don't do either or both get canned.
2. Sell-ability. Honda would like to, one day, put one of these in the homes of lazy geeks who would like to have a machine that can mow the lawn or help them move that couch downstairs... Sure the cart-bot is "easy", but people are more attracted to anthro forms. Witness the Sony dog-o-matic (or whatever it was called). People like robots because they are mechanical people. If it's not a mechanical person, it's just another machine. At least that's the public perception.
This approach probably wouldn't work for industrial robotic assembly
This is a totally different thing compared to industrial assembly. Most industrial robots are specialized, ie, they are designed to do one task or type of task and do it to exacting specifications ("don't tighten the screw too tight"). This robot is a generalist. It doesn't do anything particularly well but it does a wide variety of types of tasks. Industrial robots are like Peterbilts. They're good for one thing (driving heavy, big things long distances) but suck at anything outside of that realm (parallel parking, drag racing). This guy is more like a pickup: It can haul goods (although not as well as a Peterbilt), it can parallel park (although it needs a biggish spot) and it can drag race (although it won't win... at least it beats the Peterbilt).
Back on topic:
Okay, you seem to have a much more detailed history with robotics than I.... so maybe this is a dumb question but... what about radar or some other non-visible em wave? At the very least that would eliminate the venetian blind noise. Failing that, is a sonar-based approach a viable option?
2 1337 4 u!
Apparently you can also download the dangerwillrobinson.wav file for this thing.... but I forget the url :)
2 1337 4 u!
How much longer till they release the M-P3 battle droid? Programmed to battle big record label executives while serving copywritten music. Could happen.
Nitinol is cute, but very limited. It needs a huge power to generate a small force - most is wasted in heat.
Mondotronics have a cute project book and kit for Nitinol, which is a splendid birthday present for any geek larvae you might know. Milford Instruments sell it in the UK.
If you want a more workable muscle for small robots, look at the Air Muscle from Shadow Robots. These are interesting because they generate a pull action from air pressure, yet in a small package.
Al Gore
> many modern devices use a single laser beam to judge distances, why not this robot?
Lasers don't necessarily do all that well when teh surface in question doesn't reflect red light. Taking samples from three different bands, over expansive fields of view, allows albedo calculations off multiple points to be made in a vast array of conditions.
--Dan
How long could it be before I can get a fully loaded battle mech? Complete with rocket launchers and machine guns? Oh yeah! That'd make the commute to work fun again!
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?