Big Dog Robotic Dog/Mule

Translate This Thread From English to

Threaded View
This thing has me thinking that the field of robotics is about to take a
giant leap. It's got fantastic mechanics, and a walking/balancing
algorithm that should win its programmers a Nobel Prize. It is truly
amazing!

http://video.google.com/videoplay?docidS49770802105160028

-- Gordon

Re: Big Dog Robotic Dog/Mule


that video has been around for ?4? years...

one of the videos has footage of it JUMPING!  about
5 feet distance.


It carries 150 pounds of cargo!  Kick it and it doesn't fall over,
it runs on unstable ground (rocks, snow, uphill loose rocks)

one of the most impressive robot videos I've seen.


Rich

Re: Big Dog Robotic Dog/Mule

I may have linked to an older video, as I wanted one on a mainstream
site. There is a new version of the robot, and new videos. But here's
one of the new vids that's now on YouTube:



The payload of the new version has been increased to 340 pounds, BTW.
Even more impressive!

-- Gordon


aiiadict@gmail.com wrote:


Re: Big Dog Robotic Dog/Mule


watching it slip around on the ice is awesome...

you can hear a 2 stroke gasoline engine running...  I wonder
if they run servos on the legs (gaso is generator spinner) or
perhaps hydraulics (gaso is hydraulic pump spinner)  What
are the legs wrapped in?  perhaps they cut the legs off a real
dog, attached them, and want to cover it up so you cant see :-)

If I had one of those, I'd take it for walks and watch the little
children scream in horror :-)



Rich

Re: Big Dog Robotic Dog/Mule

    --I'm waiting for the miniature one for Xmas or the medium-sized one
for Battlebots, bwahaha! It'd make an awesome 'wheelchair' too, yes? Might
have Kamen trumped with this one!

--
        "Steamboat Ed" Haas         :  Dare me to              
        Hacking the Trailing Edge!  :  make less sense...
                          www.nmpproducts.com
                   ---Decks a-wash in a sea of words---

Re: Big Dog Robotic Dog/Mule



It's already here too.

http://www.bostondynamics.com/content/sec.php?section=LittleDog




Re: Big Dog Robotic Dog/Mule


    --Oh wow that's extremely neat! Doesn't look like it's for sale
tho.. Gotta get on the waiting list I guess.. :-(

--
        "Steamboat Ed" Haas         :  Dare me to              
        Hacking the Trailing Edge!  :  make less sense...
                          www.nmpproducts.com
                   ---Decks a-wash in a sea of words---

Re: Big Dog Robotic Dog/Mule


I was really impressed with that new work as well.

However, one thing I noticed in that video is that it seemed to be using
very different walking gaits in each segment.  My assumption is that they
are developing and testing different gait algorithms for different tasks.
That's a fine thing to do, but stringing the different tests together on
the video gives the illusion that the robot is better than it is - because
it looks like it "knows" how to deal with all those different terrains by
using different walking (or jumping) techniques.  No doubt, they still have
to work out how to combine multiple walking algorithms, all probably very
different, into one general system that can adjust to the task at hand
based on the situation.  I suspect they still have a lot of work to do to
make that happen.

--
Curt Welch                                            http://CurtWelch.Com/
curt@kcwc.com                                        http://NewsReader.Com/

Re: Big Dog Robotic Dog/Mule


    No, it looks like they have that figured out.  I suspect the
whole system is primarily gaitless - legs are viewed as an asset to be
used to accomplish a goal.  Gait is an emergent behavior.

    Step through the sequence where it slips on ice and recovers, and
pay attention to when the feet lift, when they slip and when they don't,
and the primacy of slip control over balance control over foot sequencing
over direction control.  Don't think gait; think adaptive feedforward
control.  Read Raibert's old "Legged Robots that Balance" and my
work from 1995.

    Finally, someone is doing this right.

                    John Nagle
                    Animats

Re: Big Dog Robotic Dog/Mule


And in any case, as a "gait" it's not based on the anatomy of any animal
on earth. The legs are in an unusual pushme/pullyou configuration.

It's also interesting that the legs never stop moving, even with the
robot is keeping place.

-- Gordon

Re: Big Dog Robotic Dog/Mule


Big Dog looks a little strange, but in fact, the legs on quadrupeds
are
arranged in front-back mirror-symmetry fashion.

http://www.ai.mit.edu/projects/leglab/robots/3D_hopper/3D_hopper.html

Also, Boston Dynamics has had videos of BD on its site for a couple
of
years or so.

http://www.bostondynamics.com/content/sec.php?section=BigDog

Even back then, they were experimenting with slightly different leg
arrangements with different #leg segments, as you can see in the
original video, betwen the general walking scenes and the
jumping scenes at the end.





Re: Big Dog Robotic Dog/Mule


Sorry, wrong link.

http://www.oricomtech.com/projects/leg-anat.htm




Re: Big Dog Robotic Dog/Mule


Erp, I guess it was the symmetry of the legs that threw me. Elbows and
knee-caps face one another.

-- Gordon

Re: Big Dog Robotic Dog/Mule


Humans are built the same way. It's probably been a while since
you were 2-YO, but try getting down on all fours and crawling on
the floor. Knees hit elbows. You're doing the "Big Dog".

Later on, you can graduate to the "Time Warp".  LOL.

http://blogs.chron.com/beltwayconfidential/timewarp.jpg
http://en.wikipedia.org/wiki/Time_Warp


Re: Big Dog Robotic Dog/Mule



    Big Dog is unusually symmetrical, but that's probably
a manufacturing economy.  All four legs appear to use
exactly the same parts.  That's just good design.  It's
nicely worked out; the hydraulic cylinders and hoses are
all inside the leg structures.

    Foster-Miller, which makes the TALON and SWORDS military
robots, apparently did the mechanical work.

                John Nagle


Re: Big Dog Robotic Dog/Mule


I do agree that the "gait" algorithms are far from fixed sequences.  It's
clear they have a lot of very advanced adaptive balancing and movement at
work (they aren't called Boston _Dynamics_ for nothing :)).  The examples
of them kicking the animal off balance show that nicely.  And, as you say,
the slipping on the ice is priceless in that it shows just how adaptive the
system is.

But that doesn't explain the _extreme_ difference in behavior between the
videos of it walking outside (where it never stops moving the legs and it
is constantly prancing even when standing in place), and the video of it
walking over the pile of concrete cinder blocks where it keeps 3 legs on
the ground and only moves one at a time, and the video of it hopping and
doing the jump at the end.  That very much looks to me like three
completely different locomotion algorithms.  I can't imagine that their
adaptive algorithm which keeps the feet moving would work very well at all
on that pile of concrete blocks for example where the legs could get stuck
between blocks or knock the blocks off balance when it trys to "dance" on
them.

--
Curt Welch                                            http://CurtWelch.Com/
curt@kcwc.com                                        http://NewsReader.Com/

Re: Big Dog Robotic Dog/Mule


Quadrupeds have a range of different gaits they use for different
situations, depending upon speed and terrain characteristics,
ranging from slow creep to diagonal walk to trot to canters and
a couple of types of gallop. Not all quads use the same range.

http://en.wikipedia.org/wiki/Horse_gait

The 1-legattatime is the creep gait, used widely by cats, but
not so characteristic of dogs or horses, and is the most stable
for going over rough terrain. The slower gaits differ mainly by
relative leg timing, so it's not too difficult to change over between
creep and diagonal walk or trot. The gallops OTOH are quite
different leg sequencing,

As John indicated, Raibert [honcho at Boston Dynmaics] has
been working on "dynamic" adaptive walking gaits since the
early 1980s, since his time at the MIT leg lab. Lots of gimbals
and gyros.

http://www.ai.mit.edu/projects/leglab/robots/3D_hopper/3D_hopper.html
http://www.ai.mit.edu/projects/leglab/robots/2D_biped/2D_biped.html
http://www.ai.mit.edu/projects/leglab/robots/quadruped/quadruped.html
http://www.ai.mit.edu/projects/leglab/robots/robots.html





            http://CurtWelch.Com/


Re: Big Dog Robotic Dog/Mule


This one's pretty cool, too:
http://video.google.com/videoplay?docid=-8194319473624563718
And this one skitters about so much it is also cool:
http://video.google.com/videoplay?docid=-8194319473624563718

Can you imagine the psychological impact these things might have in dark jungles
or deserts with the appropriate covering & sound
effects ? Would either start or stop you from drinking !!! :-)

JCD


Re: Big Dog Robotic Dog/Mule


Here is yet another video - showing it going uphill, downhill, and even running
jumping and kicking up its heels at the end of the
movie! COOL!

http://www.coasttocoastam.com/gen/page2490.html?theme=light  


Re: Big Dog Robotic Dog/Mule

I totally agree with you Gordon.  There will be huge changes in the next
10 years.  I Live 35 minutes from this company.  I can't even begin to
tell you how badly I want to work there.  I'd sweep the floors if I had
to.  ;-)  I hope those guys read this thread.

Shawn



Gordon McComb wrote:


Site Timeline