Slashdot Mirror
Six-legged Robot Teaches Itself To Walk
rabiddeity writes "An undergraduate at the University of Arizona has built a six legged robot from scratch. The robot, which is equipped with sensors on each foot, teaches itself to walk and orients itself via an onboard camera. A similar design might be used to explore unstable environments such as collapsed buildings or rocky landscapes."
So it looks like we only had to wait a few hours for AI to surpass the abilities of a drunken man. Can't wait until tomorrow morning.
These six-legged robots can dance! Hexapod: Best of Dance 2009
Yeah, this in no way lessens the accomplishment of a robot actually learning to walk, but I figured it was half on-topic, half cool-as-hell so I'd post it :-)
Four legs bad.
93rd rule of Slashdot: No matter how obvious my sarcasm is, my comment will be taken seriously by someone.
A similar design might be used to explore unstable environments such as collapsed buildings or rocky landscapes.
No, it won't.
I made a hexapod with 3DOF per leg that could walk in any direction "from scratch" by myself, in high school, for fun.
Adding some foot sensors is the obvious next step, and I've heard a lot about learning algorithms for walking robots being used over the years.
Honestly, I'm only bitter because I made something cooler in college but never bothered to post it online, so no one saw it aside from my classmates. But, it was a battery-powered 4 legged walking robot that ran a micro ITX windows XP pc inside its body, and was controlled through the internet with a remote PC by a wireless Xbox 360 controller.
It was honestly totally badass.
Oh, and it could support 20lbs static weight on its standard size hobby servo motors (but they were the $115 ones).
But more than anything, my point was is wasn't that hard, and that robot wasn't going to be walking through rubble any time soon.
It *could* have, but it wasn't going to. Neither is this one. People have been building basic hexapods for a long time. We still haven't sent one to the moon.
-Taylor
Worldwide Military budgets: $2100 billion. Worldwide Space Exploration budgets: $38 billion. Really, world? Really?
The exciting thing is that the robot could compensate when part of itself was damaged and get around/over obstacles
"Maybe this world is another planet's hell"
Aldous Huxley
"Stumbling around on six legs isn't very hard. "
That link didn't show what stiquito could do. Here's a video.
"2010 is a little late to be doing a six-legged crawler. They're fun to build, but you don't issue a press release."
I think parent is right, seems six-legged robots have been around forever. An electrical engineer senior shouldn't have a problem building one of these without a kit, although it looks like he might have used this kit. Sure the legs look a bit different, but the placement of servos, etc look the exact same, and before someone says "how many different ways can you build a hexapod robot?" there's many different designs
my karma will be here long after I'm gone
Isn't this the MSR-H01 Hexapod?
here's the student's video
Here's video of the MSR-H01 Hexapod:
video 1
video 2, at 1:35 it does similar "body wave" movements
The legs look different, but the student does say on that youtube description "This is a demonstration of the new leg design which is much more solid than the previous design."
my karma will be here long after I'm gone
This is similar to a stiquito in only the most superficial way -- its a movable machine with 6 legs. In every other way its different.
Controls: The stiquito has a single (or sometimes two) actuators, that are placed to mechanically, repeatedly cause the same walking motion. This student's robot has 12 actuators, 2 joints on each leg. This makes the robot much more versatile, but also makes the control problem much harder to solve.
Learning: A stiquito is dumb -- you attach the SMA to the legs, and put a current through to tighten them. It works exactly the same every time, and you have to put it together in just the right way to make it work. This robot is self-learning (or more exactly, learns through reinforcement). The designer simply creates a fairly simple algorithm that has it try motions and see if it gets it to move in the desired direction, and then learns how to do it over time.
While I think its fair to say anyone with some mechanical aptitude and knowledge of machine learning could put something like this together, its not exactly a simple feat and is certainly impressive for an undergrad. I don't know of any other self-learning six-legged robots (reflecting my ignorance only), but given the capabilities plus the (likely?) low cost its nothing to sneeze at and could have uses in things like disaster operations.
The summary does not do the article justice. This is the first line from the actual article:
The exciting thing is that the robot could compensate when part of itself was damaged and get around/over obstacles
Actually, that's part of the learning algorithms that have been around for a long time. Since it can teach itself to walk, it can re-teach itself with broken appendages.
-Taylor
Worldwide Military budgets: $2100 billion. Worldwide Space Exploration budgets: $38 billion. Really, world? Really?
Imagine a swarm of these things communicating via wireless or 3G, sending walking algorithms to each other while traversing difficult terrain.
I've built enough walking robots to not be too chilled by that vision. They'll just kinda poke along, really slowly, and then their batteries will die.
Awesome.
-Taylor
Worldwide Military budgets: $2100 billion. Worldwide Space Exploration budgets: $38 billion. Really, world? Really?
Imagine a beowulf cluster of these things communicating via wireless or 3G, sending walking algorithms to each other while traversing difficult terrain.
You seem to have made an error in your comment. I took the liberty of fixing it.
Sewage Treatment Facilities - "Our duty is clear."
That is correct. Mark Tilden has been doing similarly cool walking robots (but mostly in analog!) for years now. Check this out:
http://www.youtube.com/watch?v=gM4DitOJdyA
I remember seeing a video of one of Mark Tilden's robots (or maybe it was one of Rodney Brooks) and he was able to bend a leg back and it would keep walking successfully with the remaining legs. The beautiful part was that there was no microcontroller involved - it was simple analog circuits replicating neuron functions. The class of robotics Tilden founded is called "BEAM robotics" - more information can be found here:
http://en.wikipedia.org/wiki/BEAM_robotics
http://www.solarbotics.com/
Slashdot's first reaction to VMware
I saw this a few weeks ago when it was emailed to all the students at UA. Here is a video of the guy who made it, and it shows the robot walking around. The video mentions that IBM bought it from him. http://uanews.org/node/29644
Hexapodia is the key insight.
The world is quite different ever since the robotic uprising...
There is no more unethical treatment of the elephants.
Well, there's no more elephants, so...
Ah, but still, it's good.
Rodney Brooks did this at MIT 20 years ago.
This is news how? I'm hoping (didn't read the article) that there is something special in what they've done, cause this is old news.
http://www.cs.cmu.edu/~chuck/robotpg/attilapg/
http://people.csail.mit.edu/brooks/papers/colt.pdf
http://books.google.com/books?id=VQcCV1VuT_cC&pg=PA71&lpg=PA71&dq=mit+atilla+learns+to+walk&source=bl&ots=n9YkssitMh&sig=zYJ-SRu4KZ7IsWXTPAWeXHVMqCY&hl=en&ei=gZxzS-HeCJCI8Aahg4ydBQ&sa=X&oi=book_result&ct=result&resnum=5&ved=0CB4Q6AEwBA#v=onepage&q=&f=false
Is that it was impressive enough to catch Intel's attention. It isn't as though this guy was going around to all the news agencies saying "Hey! Look! I made a robot!" No, he made a robot that really impressed his professor. News of it somehow got back to Intel, I suspect his professor probably is friends with someone there, and they said "Wow, that is an amazing little robot. This interests us in particular since it uses our processor." Ok well when a major company is interested in something your university made, you sure as hell put out some information about it. Do remember that universities are having their budgets cut left and right. Might do some good if people were reminded that cool, commercially applicable, stuff can come from them.
Also, if all you saw was 6 legs, well you didn't look very hard. The reason Intel's interest was peaked was the legs, it was how it works. That Stiquito is a simple device, probably a finite state machine, that just does the same thing over and over. Notice that what it has no sensors, just an on/off switch. You turn it on, it follows whatever program is in there to move forward. Not the case with this thing, it uses its camera to see what is happening, and then figures out what to do. It is actually processing data and adapting based on that. Much, much more complex.
He built it for a cognitive robotics class, so the emphasis was on the software, not the hardware (it uses a webcam and optical flow calculation for movement detection, for feedback into the learning algorithm). The FOX article is a horrible source for this story, but if you Google a bit you can find that he used a 3-D printer to build his own legs for the slick version shown - definitely not a kit!