Robot Solves Rubik's Cube In Less Than a Second (livescience.com) 54
An anonymous reader quotes a report from LiveScience: In just over half of a second (0.637 seconds), the Sub1 Reloaded robot made each side of the Rubik's Cube show a single color. This breaks the previous record of 0.887 seconds achieved by an earlier version of the same machine using a different processor. German technology company Infineon staged the record attempt at the Electronica trade fair in Munich this week, as a way to highlight its self-driving-car technology. The company provided one of the Sub1 Reloaded robot's microchips. Infineon said more than 43 quintillion combinations of the Rubik's Cube's colored squares are possible. That same number of cubes would cover Earth in 275 layers, resulting in an approximately 65.6-foot-high (20 meters) layer of Rubik's Cubes, the company added. The record-breaking attempt began with the press of a button. Sensor cameras on the machine had their shutters removed, and the computer was then able to detect how the cube was scrambled. The computing chip, or the "brain" of the machine as Infineon called it, then determined the fastest solution. Commands to execute the solution were sent to six motor-controlled arms. "It takes tremendous computing power to solve such a highly complex puzzle with a machine," Infineon said in a statement. "In the case of 'Sub1 Reloaded,' the power for motor control was supplied by a microcontroller from Infineon's AURIX family, similar to the one used in driver assistance systems."
Re: (Score:2)
it just applied a new set of colored stickers to each side to make it appear to have been solved!
Huh? (Score:3)
Re:Huh? (Score:5, Interesting)
How is solving a rubik's cube ANYTHING like self driving?
It isn't. Solving Rubiks Cube is trivial [rubiks.com]. Anybody can learn to do it, and many people can solve a randomized cube in under a minute.
A computer can find the solution in a few microseconds. The hard part isn't finding the solution in software, but building a mechanical contraption to rapidly twist the cube without breaking it. This is an achievement in mechanical engineering, not software. TFA completely skipped over the substance to focus on the trivial.
Re: (Score:2)
Re:Huh? (Score:4, Interesting)
Re: (Score:2)
In the video the cube did look like it might have had some modification. There are significant bevels on the corners, presumably so the alignment can be off by more and it will self-correct as you twist it.
Re: (Score:1)
Looks like a Dayan Zhanchi stickerless cube or a clone of one.
They are pretty common speed solving cubes.
Re:Huh? (Score:4, Interesting)
The hard part isn't finding the solution in software, but building a mechanical contraption to rapidly twist the cube without breaking it.
Not just building it, but controlling it. Both are hard problems, and both are limiting factors.
Re: (Score:2)
I wouldn't say the solving part is trivial. It looks like it's solving it in around 20 moves, very close to the theoretical minimum. The solution you linked to doesn't come anywhere near that efficiency, it already needs well over 20 moves just for the first layer. Coming up with extremely short solutions does require an enormous amount of computing.
I do agree that it's nothing like driving a car: those AIs use neural networks that look for approximate solutions based on fuzzy data. Completely different fro
This is obviously old news (Score:1)
If indeed "this breaks the previous record of 0.887 seconds achieved by an earlier version", then the headline is old news.
Does it cheat? (Score:3)
I remember last time this machine set the record, there was some debate as to whether it should count, as the cube has to be modified in order to be mounted in it. The robot doesn't grasp the cube, but rather its six arms have pins that are inserted into holes drilled in the center square of each side.
Apples To Apples (Score:2)
Human did it in 4.74 seconds 5 days ago (Score:5, Informative)
https://www.youtube.com/watch?v=tLksISrKtO8 [youtube.com]
Re: (Score:2)
Although that doesn't include the time he was allowed to examine it before starting. Here's the video:
https://www.youtube.com/watch?v=tLksISrKtO8 [youtube.com]
Yeah, that "tremendous computing power" quote doesn't sound right. Humans are quite able to figure our the algorithm in a few seconds, I don't see why a computer couldn't use a similar algorithm to get the answer while iterating far fewer than "43 quintillion combinations", quickly determining the sequence of moves should be the easiest and quickest part.
The more impressive part is the hardware side, making a custom machine that doesn't tear the cube apart. But unless I'm missing something this strikes me a
Re: (Score:2)
Because the people that programmed this thing have never read any of the books written on solving a rubik's cube. There is *ONE* solution; one sequence of moves that when repeated will eventually solve the puzzle. There's no need to think out a solution. Simply pick up the cube and start repeating the pattern until all the sides match. (btw, that's how real people do it.)
Re:Human did it in 4.74 seconds 5 days ago (Score:4, Informative)
Because the people that programmed this thing have never read any of the books written on solving a rubik's cube. There is *ONE* solution; one sequence of moves that when repeated will eventually solve the puzzle. There's no need to think out a solution. Simply pick up the cube and start repeating the pattern until all the sides match. (btw, that's how real people do it.)
No such solution exists. The best you can do by repeating the same pattern is to cycle through 1,260 states, multiplied by the length of the permutation sequence: https://people.kth.se/~boij/ka... [people.kth.se]
There does exist at least one sequence of moves that is guaranteed to solve the cube, eventually (it forms a Hamiltonian circuit.) It's 43 quintillion moves long, and you can download a (200MB) specification describing how to construct the sequence here: http://bruce.cubing.net/ham333... [cubing.net] Iterating a sequence that long is well beyond the capabilities of both human and robot, sadly.
Modern solvers use variations of the Kociemba algorithm, which can find near-optimal solutions very quickly: http://kociemba.org/cube.htm [kociemba.org] CPU power is important because more time spent searching can yield shorter move sequences - the slowest part of the solve (computer vision, solution search, twisting the cube) is the physical part. However, every millisecond spent searching for shorter sequences might be better spent actually executing a suboptimal solution.
Re: (Score:2)
while iterating far fewer than "43 quintillion combinations"
Solving a Rubik's cube doesn't require iteration at all. There's a logical process you can follow without any guess work.
Re: (Score:1)
Custom Rubik's Cube? (Score:3)
Re: (Score:2)
Exactly. This doesn't impress me. Let me see a robot with arms that picks up a Rubik's cube rather than a custom construction specifically made for this robot.
43 quintillion? (Score:2)
Re:43 quintillion? (Score:5, Interesting)
Apparently it still works out to be that large, according to this [wikipedia.org]:
There are 8! (40,320) ways to arrange the corner cubes. Seven can be oriented independently, and the orientation of the eighth depends on the preceding seven, giving 3^7 (2,187) possibilities. There are 12! / 2 (239,500,800) ways to arrange the edges, restricted from 12! because edges must be in an even permutation exactly when the corners are. [...] Eleven edges can be flipped independently, with the flip of the twelfth depending on the preceding ones, giving 2^11 (2,048) possibilities.
8! × 37 × (12! / 2) × 2^11 = 43,252,003,274,489,856,000
Including all permutations is about 12 times that, around 519 quintillion.
Re:What does it mean to 'solve a rubiks cube'? (Score:5, Informative)
Solves -which- permutation of a Rubik's cube in less than a second? Every single one? How can they prove that before the end of the universe?
It has been proven [cube20.org] that 20 moves suffice to solve Rubik's Cube from any starting position.
If you restrict each move to a quarter turn, then 26 moves suffice.
The proof only took 35 years of CPU time.
Is it smoking? (Score:2)
I'd be more worried about the damn cube flying apart with these increasing speeds, not the CPU behind it. Overshoot a rotation and then start the next one before you correct it, and the cube will explode. Humans will not generate the force necessary to break a healthy cube, but even then they still sometimes come apart under these conditions (without breaking, the center pieces are spring-loaded).
Newsflash: Machines faster than humans! (Score:5, Funny)
Awwww... (Score:1)
Congratulations! (Score:2, Interesting)
They're making the cube sound harder than it is (Score:2)
They're making the cube sound harder than it is. The difficulty doesn't correspond well to the number of combinations. Back in the 80s when I played with them, the solution technique I knew was based on recognizing that the components of a cube could be flipped or twisted, with the flip or twist balanced out by another component. Then you simply executed moves to undo the flip or twist. My best times were 3 minutes or so, which sucks now but I bet the solution algorithms have gotten way more sophisticat
Not a usual cube (Score:2)
That's nothing (Score:1)
Eat my shit, Infineon (Score:1)
"It takes tremendous computing power to solve such a highly complex puzzle with a machine" BOLLOCKS! I wrote a BASIC program to solve Rubik's cube on a Sharp PC-1500 pocket computer back in 1982 (and I was a teenager!). It was less than 3K (I had a RAM expansion module) and ran happily on an 8-bit Z80 CPU. Input the cube status data, wait, follow the printed instructions. Tremendous computing power my ass.