[Unofficial] CubeStormer II - robot solves Rubik's Cube faster than human world record

[IAssemble]

How long did it took you to build it? And after building it, how long until the first successful solve?

It is very difficult to answer these questions...

I think Mike and I started working together on this project about 8 months ago although he had been working on the mechanical design before this. Many of the software and construction ideas were developed as part of my previous "Android Speedcuber" and Mike's previous "CubeStormer" creations so overall it has been several years since we started.

I don't recall exactly when CubeStormer II solved it's first cube but I do remember the feeling of excitement when it did!

It's taken a long time but it's been great fun!

 

[Phlippieskezer]

That is awesome.

Much congratulations. I shall now revere you forever.

 

[cubeflip]

GIVE THE ROBOT INSPECTION!!

But I still love it.

 

[IAssemble]

GIVE THE ROBOT INSPECTION!!

But I still love it.

Or take away human inspection?

Glad you like it!

 

[souljahsu]

I am inspired.

 

[asportking]

And with a storebought as well? Very impressive.

 

[IAssemble]

I am inspired.

Thanks, that's great to hear

 

[EeeeeWarne]

Give it a ZhanChi and we go sub 5!

By we, I mean human kind...since it was humans who made the robot...

 

[JyH]

Brest

 

[IAssemble]

Give it a ZhanChi and we go sub 5!

By we, I mean human kind...since it was humans who made the robot...

Well if you are thinking in human terms... then we could just apply human timing rules and exclude the inspection time in "which case we already have"

Or are you thinking BLD sub5?

 

[Lucas Garron]

I see *someone*'s been keeping track of things.

...a solution of around 21 moves optimised specifically for the mechanics of the robot...

Finally! Care to elaborate on the details? (Different pruning, different metric, generate a bunch of solutions and filter by robot metric?)

Also, see this for more ideas I want to see in a cube bot.

 

[Enter]

wow that is cool

 

[Kirjava]

No, it's well and truly British!

omg

will you bring it to UK Open?

 

[~Adam~]

Amazing. I would probably buy one just to scramble cubes that fast.

 

[Schmidt]

Or are you thinking BLD sub5?

Yeah. Put a blindfold on the camera and see if it can remember those 21 moves!

By "comparable with a human" you mean mechanically comparable by forcing it to do roughly the same number of moves?

In that case, would it be allowed to exclude the "inspection time too"?

Thanks, I'm glad you like it

Just give it 15 sec of inspection instead of 50-200ms. And force it to do white cross, corners, second layer, 2-look OLL and 2-look PLL and see if it can beat Erik

 

[IAssemble]

I see *someone*'s been keeping track of things.



Finally! Care to elaborate on the details? (Different pruning, different metric, generate a bunch of solutions and filter by robot metric?)

Also, see this for more ideas I want to see in a cube bot.

Hi Lucas, I already responded to this with this

But yes, "generate a bunch of solutions and filter by robot metric" sums up most of the changes.

 

[Sillas]

great job. this robot beat Feliks.

 

[IAssemble]

Yeah. Put a blindfold on the camera and see if it can remember those 21 moves!

haha - by BLD I was referring to the fact that BLD times include the inspection time as part of the total time so a sub5 total time would be a cool target given the current total time shown time in the video.

Just give it 15 sec of inspection instead of 50-200ms. And force it to do white cross, corners, second layer, 2-look OLL and 2-look PLL and see if it can beat Erik

But what's the point in giving it 15 seconds of time if you want to force it to solve like this.... it could probably calculate a cross, corners, second layer, 2-look OLL and 2-look PLL solution in a few hundred microseconds?!?

But Erik's physical manipulation skills are awesome!

 

[Stefan]

But what's the point in giving it 15 seconds of time if you want to force it to solve like this.... it could probably calculate a cross, corners, second layer, 2-look OLL and 2-look PLL solution in a few hundred microseconds?!?

It could use the time to find short/fast solutions. An optimal non-lucky one would be nice.

 

[IAssemble]

It could use the time to find short/fast solutions. An optimal non-lucky one would be nice.

I'm confused by these last few comments.... doesn't searching for a short/fast solution conflict with the idea of forcing it to use a human algorithm? Humans don't have enough capacity to search very much (I assume - after all I don't actually know how Feliks Zemdegs' brain works! ) so why let the robot search if you're trying to compare like-for-like?

CubeStormer II manages about 5 or 6 moves per second whilst human speedcubers manage about 9!?! (Feliks - 51? moves in 5.66s) So if the robot is artificially constrained to use human solutions then the human will win...