[youtube=http://www.youtube.com/watch?v=O3ovrT8pWww]
This tip was submitted by [Mike], with the original information seen in this post. When I passed the story along to our writer [Mike Szczys] I didn’t send along the entire email conversation. This bot is noteworthy because it has taught itself to walk. In the build log you can learn about how it has created its gait and altered it based off of the vision. There are also some great pictures of prototypes there too. While we can all agree that it isn’t as impressive looking initially as the A-Pod, remember that it wasn’t programmed to look impressive.
Wow……just… Wow!!!!!
HAD has been HAD!!!
Suddenly a lot more impressive.
Ok, I should have RTFA – it taught itself to walk!
Nice work…that is brilliant.
this went from being “cool” to being SWEET!!! now that it is known that it taught it self to walk
Pretty amazing how one minor detail makes this a hohum project into something pretty cool.
I was thinking that an Atom was overkill. Now I can conclude that it is not.
Wow.
I thought the thing was really impressive without realizing it taught itself to walk. Now I am blown away.
I was wondering before (in the previous post) why this was so impressive, so I read the article. The fact that it uses the camera to help teach itself is also very cool. I could see NASA using something like this (self teaching hexapod) to explore Mars (and replace the current rovers). Mars doesn’t really work well for wheeled rovers because of all the rocks, but a bigger version of this could possibly even RUN on the surface. The fact that it can teach itself how to walk means that the NASA folks don’t have to do as much work to make sure it’ll be able to walk.
Now, when can we hear a robot teach itself to speak/sing? Hopefully not like GLaDOS/HAL.
Oh wow, it learned how to do that?? That is exceptionally cool. Now all we need are 70 foot tall models that are sold to the public on a regular basis.
@cryozap
While I admit that legs seem better than wheels, the mars rovers’ original 90 martian day missions are up over 2000 martian days now.
One of Spirit’s motors failed on the 779th day and hasn’t worked since, but it still was able to move until 1165 martian days later, when it broke through rock and fell into a sand trap.
If a hexapod has a motor failure or two, what’s it going to do? You could add more legs and make them detachable, etc., but I can’t see how a set of legs would last longer than the wheels that allowed the Spirit to move for more than 21 times the intended length of time.
A shorter and much more humorous explanation is available at http://xckd.com/695/ .
Also, what’s wrong with GLaDOS? :(
Seems pretty cool if they made it look a bit nicer I’d bet it could sell.
“Here’s what I think of your puny human imperial measurements!” *stomp* *stomp* *stomp*
@RoboGuy
Redundancies for learning is the idea.
Since the hexapod is able to learn to begin with, then the hexapod is able to re-learn a new gait in the event of a motor failure. Even if 10 motors failed, the other 8 could still move.
Pre-programmed algorithms would certainly fail if a single motor would fail, but learning allows relearning of complex new geometric configurations. In other words, the hexapod walks around, a rock can crush two legs, and then the hexapod relearns how to walk with its new configuration.
So then why haven’t we integrated this already? Learning takes time and energy for experimentation and even in simulation. Current learning techniques certainly need to be made more efficient.
I don’t really understand how this thing learned itself to walk… Can anyone explain it to me? Also, the learning algorithm, how would it work?