# Gleaming The CubeMarch 20, 2018 9:20 AM   Subscribe

A Rubik's cube being solved in 0.38 seconds.

Cube bots previously on MeFi:
2017 (0.637 s)
2011 (5.352 s)
2010 (? s, but runs on a Nokia N95)
2008
2006 (~36 s - the video is quite something.)
2002
posted by zamboni (15 comments total) 17 users marked this as a favorite

Holy smokes!
posted by Big Al 8000 at 9:46 AM on March 20, 2018

RuBot II, I Love You
posted by chavenet at 10:02 AM on March 20, 2018 [1 favorite]

I'm increasingly confused about these speed records. I'd hazard a guess the limiting factor here has more to do with mechanics (the purpose-built something designed to rotate the cube as fast as possible) than with challenge of determining a "solution" (the program that takes the scanned cube faces and calculates the minimum twists and turns). The latter must now happen much faster than the 380 ms it takes to manipulate the physical cube.
posted by Insert Clever Name Here at 10:16 AM on March 20, 2018

Yeah, moving the motors at those speeds, in synch, without destroying the cube (or the motors) is the impressive part.

In the 80s there was a small book on Rubik's Cube that consisted of recognizing patterns and applying particular patterns of movements to solve them. The logic would be absolutely trivial for software, and color recognition not much more difficult.
posted by Foosnark at 10:28 AM on March 20, 2018 [1 favorite]

The vision's also a bit tricky, but apparently not too hard. The only special thing they did is use Playstation game console cameras because they were lower latency than standard webcams.

The article goes into detail about the mechanical problem. Liked the insight that stiffer cubes are better for the robot because it is precise and wants precision movement.
posted by Nelson at 10:37 AM on March 20, 2018 [1 favorite]

Yeah, solving it is a homework problem for a slightly old-fashioned AI class.

What I don't understand is why you'd need vision at all, beyond an initial picture of the unsolved cube. The robot knows what it did, so it knows how it looks at every point...
posted by kleinsteradikaleminderheit at 10:43 AM on March 20, 2018 [1 favorite]

> The robot knows what it did, so it knows how it looks at every point.

Yeah that crossed my mind too. On the other hand, it needs the cameras for the initial scan, and it can probably benefit from validation that a recent move resolved and did not cause the cube to explode.

(Seriously, everybody make sure to watch the "Cubesplosion" video, it's a thing of slo-mo beauty.)
posted by ardgedee at 10:54 AM on March 20, 2018 [1 favorite]

calculates the minimum twists and turns

On that: I noticed in the slowed-down video that one of the moves is a double: the central slice is held still while both the top and bottom slices both rotate.

That looks like a move that'd be hard for a human to do fast -- might have to be done as two sequential moves? -- but which their robot can perform efficiently by rotating the two slices concurrently. It makes me wonder if there's scope for further reducing their time by using an algorithm that favors these robot-friendly moves.
posted by We had a deal, Kyle at 12:16 PM on March 20, 2018 [4 favorites]

Elicited a whispered "daaang" from yrs truly.
posted by potrzebie at 2:23 PM on March 20, 2018

It looks like they compute the solution first: “Once the faces are identified, the solve sequence is computed using the min2phase algorithm. This returns a solution which we've observed to be 19-23 moves for a well-scrambled cube.”

Which makes a lot of sense, but then it’s not clear how much advantage the low latency cameras give. But at this speed every millisecond counts.
posted by sjswitzer at 6:54 PM on March 20, 2018

I like the video with a Nokia, because there's a trance soundtrack and a montage. It's ridiculous and I am here for it.
posted by Pronoiac at 7:36 PM on March 20, 2018

On that: I noticed in the slowed-down video that one of the moves is a double: the central slice is held still while both the top and bottom slices both rotate.

That looks like a move that'd be hard for a human to do fast -- might have to be done as two sequential moves?

It's just a single rotation of the central slice, in essence, which is probably how a human would do it. With the cube being held on all six faces, rotating the outer faces around the central slice is the only way the machine can do it, but it's completely equivalent - the middle slice rotates relative to the outer ones
posted by Dysk at 7:30 AM on March 21, 2018 [2 favorites]

Also this is super cool and the first video made my boyfriend and I burst out laughing in sheer delight. Have watched the frantic "BZZZT" of real time solving like a dozen times now.
posted by Dysk at 7:32 AM on March 21, 2018

Not possible. The faces of the cube would have to move faster than c. Fake.
posted by quillbreaker at 5:24 PM on March 21, 2018

I don't think we can say the cubes are being solved if we haven’t done a root cause analysis to figure out why the cubes are getting jumbled up in the first place.
posted by um at 6:45 PM on March 21, 2018 [4 favorites]

« Older Metafilter.com: You've thought about a lot of...   |   What Is Democratic Socialism Anyway? Newer »