Lego Robot Solves Rubik's Cube Puzzle In 3.253 Seconds

SternisheFan sends this news from CTV: "The Cubestormer 3 took 18 months to build but only needed 3.253 seconds to solve [a Rubik's cube], breaking the existing record. Unveiled at the Big Bang Fair in Birmingham, U.K., the Cubestormer 3 is constructed from the modular children's building-block toy but uses a Samsung Galaxy S4 smartphone with a special ARM chip addition as its brain. It analyzes the muddled-up Rubik's Cube and powers each of the robot's four 'hands,' which spin the cube until all sides are in order. Created by ARM engineer David Gilday and Securi-Plex security systems engineer Mike Dobson, Cubestormer 3's new record shaves just over two seconds off the existing record, set by Cubestormer 2, which the pair also built."

Comments

alas !

[invictusvoyd]

Only if I had four hands ..

Re:alas !

[ls671]

Well, I guess it also depends on "how well" the cube is when you start.

Is there anybody that knows the longest possible sequence of move one would have to do in order to resolve the cube? In order words: what is the worse configuration to start with when you try to resolve it.

I think our robot could spend more than ~3 seconds resolving it with worse configurations.

Now: Let's design a robot to mix the cube for the other robot and have a data set, not just one run.

Re:alas !

[The boojum]

"God's number" is 20.

Re:alas !

[Imsdal]

As others have pointed out, the optimal solution is never more than 20 moves, and this instance required a bit more than 20 moves, so it probably wasn't the very easiest of configurations. That said, I don't understand why this record isn't for "best average of 20 runs" (or some other suitable number). It wouldn't take more than five minutes to run it, and it would be a lot more telling about the actual capacity of the robot.

It would also be interesting to see the variance of the solving times. How consistent is this thing?

Re:SloMo required!

[AGMW]

Please someone upload an ultra slow motion video of the process!!!

That was the slow motion video!

They can probably compute a just-over-20-move solu

[ToastyKen]

As The boojum said, you can theoretically solve a cube in any configuration in 20 moves, but it might take a long time to find that optimal solution. Computers can quickly find a just-over-20-move not-quite-optimal solution, though.

Human-friendly algorithms generally take over 50 moves, with the absolute best solves still taking more than 40.

I stepped through this video frame by frame. They rotate the cube 5 times to inspect each face first (I guess they only have one camera), paused about 0.2 seconds (presumably to calculate a solution) and then they made 21 moves plus 4 rotations to solve it. (The rotations were necessary because it only has 4 arms and can't spin the top and bottom layers.)

So yeah, looks like they computed a nearly-optimal solution, and I imagine they can probably get a just-over-20-move solution every time.

Re:Impressed yet disappointed with Lego

[jimbolauski]

The first thing you need to know when bonding anything is the material you are bonding, most Lego bricks are made of ABS plastic and I would assume the ones you had are as well. The way you bond ABS is by chemically welding it, you can find the solvent glue at most hardware stores (plumbing section). The glue will be black because there will be abs filler in there to fill voids because it is used in plumbing and must be water tight. The second option is to use MEK (methyl ethyl ketone) to chemically weld it, MEK is clear and has no filler. Most hardware stores don't sell MEK anymore but a MEK alternative that may not work. MEK is some nasty stuff wear gloves when you use it and work in a well ventilated area with a respirator, if you get any on your gloves change them as the MEK will be eating through them shortly.

Re:They can probably compute a just-over-20-move s

[newcastlejon]

Is it really necessary to scan every face? I'm no expert when it comes to Rubix cubes (I've only ever managed two sides on mine), but isn't a cube similar to a dice where you can look at one side and infer what state the opposite face is in? When the previous machine was heralded here on /. I recall that the Mindstorms robot only looked at 3(?) of the faces. That seems like a while ago so there's a good chance I'm mistaken about that.

Hah

[rossdee]

Try reconfiguring (ie take the stickers off and put them back in diferent places so there is no solution) and then see how the robot does...

Re:They can probably compute a just-over-20-move s

[kannibal_klown]

Nope, you can have 1 face look a specific way but the "back" face look any number of ways. Heck you can solve for multiple sides and have the rest of a sides still be a mess.

Re:They can probably compute a just-over-20-move s

[tlhIngan]

I stepped through this video frame by frame. They rotate the cube 5 times to inspect each face first (I guess they only have one camera), paused about 0.2 seconds (presumably to calculate a solution) and then they made 21 moves plus 4 rotations to solve it. (The rotations were necessary because it only has 4 arms and can't spin the top and bottom layers.)

Yes, they only have one camera - it's on the smartphone they're using. When you start the machine, the smartphone platform rises, exposing the camera at which point the app snaps photos of the faces.

After snapping photos of all 6 sides, it slides the platform back while calculating the solution (it takes up to 50ms), then it just sends Bluetooth commands to the motors to spin each face.

Re:Applications

[ShanghaiBill]

what are the real applications for such an achievement?

Impressing chicks. When it comes to meeting girls, lego robotics is right up there with model railroading.