2 step finish for Roux Edges

[cubacca1972]

All algs you have for UF flipped are terrible btw.

I am assuming that you applied x or x' as appropriate to convert the F and B moves into U moves, thereby converting the algorithms into U M R algorithms.

I love <R,U,M> algorithms. I don't like regrips.

A random thought-

You could do r or r' at the beginning of the offending algorithms, then proceed with the RUM moves. You would then have to re-align R and M at the end. 50% of the time M will be aligned with either the L or R layer, so R, R', r or r' will solve the cube.

 

[Kirjava]

Then maybe the algs wouldn't be terrible!

It might be worth changing them~

 

[cubacca1972]

I did a quick check of all the cases to see if recognition was possible without peeking at the D facelet of DF.

For all the cases where UF is oriented, and the remaining M edges are flipped, it is possible to determine the case by looking at only the facelets on the U and F faces.

For all the cases where UF is flipped, only 4 of the 12 algorithms can be identified without peeking at the D facelet of DF. For each M edge at UF, there are 3 cases, one of which has a unique pattern of facelets on the U and F faces. The other 2 cases look identical, thus requiring a peek at the D facelet of DF.

 

[oll+phase+sync]

I once did some calculations on Move count for Roux
EO: 6,2 Moves - only optimal M/U turns with fixed Centers,
UL/UR: 5,4 Moves
Medges 4 Moves

But there are easy otimisations like:
- M or M' as last move in EO
- do M/M' before EO
- Using Shorter Orientation sequences like: R' U r U M U' r' U' r or F2 M' F2 U M


So I think both ideas are really competitive.
And for Corners First or NonMatchingColorBlocks this may be even beter suited.

What I don't like is looking at "invisible" Stickers - any suggestions to fix this?


P.S: I'm alsointerested in Ming Thai's 'winning solution' solution to this.

 

[Kenneth]

I tried a mostly intuitive way for the edges some time ago, you orient, pair up and permute:

1; orient one edge (any) at BD while orienting the centres (M2 away is ok)
2; EO (5 cases)
3; put the the edge that is opposite to the first in BD at FD (make a pair) but move both to U in the last M turn ( M' U2 M')
4; look at the edge that now is in FD, if the opposite is in BD it is a skip but if it is at UR or UL then move it to UB and do M U2 M/M' (you can also look directly at UR/UL, if it is a pair you got the skip, else not)
5; permute, first solve UR/UL (or UF/UB) using M2 U/U', and then use the usual Roux algs for the M-slice, AUF if needed.

I don't say this is better than anything else but it is really easy to learn, well suited for those who are new to Roux.

If you are advanced you can do 2+3 in one go using my [wiki]L5EOP[/wiki] algs.

Some examples:

Darn! I add it later, the bot that provides roux scrambles in the swedish chatroom is down and Tomas who is hosting it is on his way from Euro...

Edit: Mega scrambler! =)

I fix it soon...

M' U M' U' M2 U2 M U2 M' U M U2 M U' M2

1; y2
2-3; M' U M'
4; U M U2 M
5; U' M2 ... U M2 ... AUF

Easy =)

U' M' U2 M U2 M' U2 M' U' M' U M U M' U'

1; U2 M'
2; M' U M U M' U M
3; U M' U2 M'
4; skip
5; U M2 ... AUF

Normal up to the easy permutation.

M2 U' M2 U2 M2 U M2 U M2 U M2 U2 M' U M2 ... weird scramble

1; M'
2-3; M' U' M'
4; skip
5; This case is important to know, the UF/UB pair is sitting in position but swapped, fix that first using M2 U2 M2
5 cont; U M2 ... AUF

lolstart

One more : U' M' U M' U' M' U2 M U' M U' M' U M U'

1; M
2; U M' U M
3; U M' U2 M'
4; skipped
5; U M' U2 M2 U2 M ... AUF

 

[Kenneth]

I have been playing around with variations of my approach (solve centres and BD + [wiki]L5E[/wiki]) the latest days and now I came to this advanced style:

1; pair up BD with either the B or the D centre in U and place it into position
2; do partial EO to reach the 3+1 case (4 cases M' U M ). 3+1 is 25% of the times so it is not lucky to skip this.
3; do the rest, the orientation is fixed so only the permutation counts, totally it is 60 cases.

If you skip EO before 2, then switch from this metod and go for permutations.

An effective method but recognition for the last step sucks.

Using the same scrambles as in the last post:

M' U M' U' M2 U2 M U2 M' U M U2 M U' M2

1; U M2 U' M2
2; U
3; M2 U M U2 M' U M U M (9f*) I don't not know many cases so I cube explored them, here I found a nice one =)

U' M' U2 M U2 M' U2 M' U' M' U M U M' U'

1; M U M2
2; U M' U M U'
3; M' U' M' U M U2 M' U M2 (9f*) Nice find again =)

M2 U' M2 U2 M2 U M2 U M2 U M2 U2 M' U M2

1; M' U M U M'
2; U
3; pure flip + H-PLL, easy to recog : U F U' F' M2 F U F' M U M (11f*) Strange alg for MU but it is optimal (alt alg : U2 M U2 M U' F2 M' F2 M2 U' M )

Pure flip is one of the cases I know; U' M2 U M U M U M U M'

U' M' U M' U' M' U2 M U' M U' M' U M U'

1; U M' U' M2
2; M' U M U'
3; M' U' M' U M' U2 M U M2 [U2] (10f*) ... ok one =)

Yes, I know, the recog ruins this idea so it is not worth learning all the cases...

 

[cubacca1972]

I have been playing around with variations of my approach (solve centres and BD + [wiki]L5E[/wiki]) the latest days and now I came to this advanced style:

1; pair up BD with either the B or the D centre in U and place it into position
2; do partial EO to reach the 3+1 case (4 cases M' U M ). 3+1 is 25% of the times so it is not lucky to skip this.
3; do the rest, the orientation is fixed so only the permutation counts, totally it is 60 cases.

If you skip EO before 2, then switch from this metod and go for permutations.

An effective method but recognition for the last step sucks.

Using the same scrambles as in the last post:

M' U M' U' M2 U2 M U2 M' U M U2 M U' M2

1; U M2 U' M2
2; U
3; M2 U M U2 M' U M U M (9f*) I don't not know many cases so I cube explored them, here I found a nice one =)

U' M' U2 M U2 M' U2 M' U' M' U M U M' U'

1; M U M2
2; U M' U M U'
3; M' U' M' U M U2 M' U M2 (9f*) Nice find again =)

M2 U' M2 U2 M2 U M2 U M2 U M2 U2 M' U M2

1; M' U M U M'
2; U
3; pure flip + H-PLL, easy to recog : U F U' F' M2 F U F' M U M (11f*) Strange alg for MU but it is optimal (alt alg : U2 M U2 M U' F2 M' F2 M2 U' M )

Pure flip is one of the cases I know; U' M2 U M U M U M U M'

U' M' U M' U' M' U2 M U' M U' M' U M U'

1; U M' U' M2
2; M' U M U'
3; M' U' M' U M' U2 M U M2 [U2] (10f*) ... ok one =)

Yes, I know, the recog ruins this idea so it is not worth learning all the cases...

Recognition is one of those make or break aspects of any system.

The goals for my system were to create a two look LSE, and to have a reasonable number of algorithms to learn, and still preserve a good move count.

Why 2 looks? My premise is that fewer steps means less time lost on pattern recognition. Look. Recognize Pattern. Recall the correct algorithm. (Adjust face or slice to prepare for the appropriate algorithm) Execute Algorithm. Next Step. I figure that the fewer times you have to go through this routine, the better.

It is important for any system to have a reasonable number of algorithms. If you have solved up to the LSE step using Roux, you already need 42 algorithms to solve the last 4 corners in one look. My system clocks in at 24 algorithms (plus a few special cases, which end up finishing with the well known Permute the M slice algorithms), assuming that you are (and should be) doing the fist step intuitively.

You could simply solve UL and UR, leaving M to solve in 1 step, requiring 40 algorithms (total algs can be reduced through excellent recognition skills, and with y2, but that could get ugly).

Or, you could solve DF and DB, and solve the U layer edges with 29 algorithms. If you stick with U and M moves, you end up with several algorithms with 13 or more moves.

From a pragmatic standpoint, it is very challenging to create usable solving systems. The gold standard seems to be few steps, few algorithms, and few total moves. Usually, you end up breaking at least one of those standards.

 

[Athefre]

I like the idea. The only negative I see is that the move count seems to be a little higher than the original way. Even if the average for a comfortable set of sequences for Step 2 was 9.5 moves, could Step 1 be done in 4 moves on average? But a big positive I see is that it removes Step 4c.

 

[cubacca1972]

I like the idea. The only negative I see is that the move count seems to be a little higher than the original way. Even if the average for a comfortable set of sequences for Step 2 was 9.5 moves, could Step 1 be done in 4 moves on average? But a big positive I see is that it removes Step 4c.

It really hangs on the average move count for step 1. There are 36 possible cases (one is the endpoint of step 1, so 35). One of those 35 cases is both UL and UR solved, leaving orientation of M slice and Permutation of M, with a remote chance that M is already oriented, and an extremely remote chance that M is solved.

I have not searched all of the cases, but there seem to be several cases that are 3 to 4 moves long, so an average move count between 4 and 5 moves seems plausible.

The 35 cases are a reduction of the actual number of cases, which is 120 possible positions, assuming that the U corners are aligned. I did my reduction such that UL is either in the UL, DF, or UR position, so adjusting the M slice would be required to set up most of the cases.

On the other hand, there is no reason that the U corners need to start in the aligned position, and no real need to set up the 35 cases exactly either. That means that lots of the actual positions could be half solved already. The true number of cases would then be 12 orientations/positions for UL, times 10 orientations/positions for UR, times 4 positions for the U layer. 12x10x4=480.

For some reason, I am not that keen on finding optimal algorithms for all 480 (479) cases to get an accurate average number of moves.

 

[cubacca1972]

I have decided to name this little system WLSE, which stands for Winter Last Six Edges. I can now add it to the Winter Variation, as part of my collection of obscure alternative steps to existing systems.

Now I have to decide between refining WLSE some more, or working on WOrtega, WWaterman, or WGuimond.

 

[Kenneth]

Recognition is one of those make or break aspects of any system.

The goals for my system were to create a two look LSE, and to have a reasonable number of algorithms to learn, and still preserve a good move count.

Why 2 looks? My premise is that fewer steps means less time lost on pattern recognition. Look. Recognize Pattern. Recall the correct algorithm. (Adjust face or slice to prepare for the appropriate algorithm) Execute Algorithm. Next Step. I figure that the fewer times you have to go through this routine, the better.

It is important for any system to have a reasonable number of algorithms. If you have solved up to the LSE step using Roux, you already need 42 algorithms to solve the last 4 corners in one look. My system clocks in at 24 algorithms (plus a few special cases, which end up finishing with the well known Permute the M slice algorithms), assuming that you are (and should be) doing the fist step intuitively.

You could simply solve UL and UR, leaving M to solve in 1 step, requiring 40 algorithms (total algs can be reduced through excellent recognition skills, and with y2, but that could get ugly).

Or, you could solve DF and DB, and solve the U layer edges with 29 algorithms. If you stick with U and M moves, you end up with several algorithms with 13 or more moves.

From a pragmatic standpoint, it is very challenging to create usable solving systems. The gold standard seems to be few steps, few algorithms, and few total moves. Usually, you end up breaking at least one of those standards.

Ops, I missed that one...

Yes, I usally break the easy recognition in favour for short algs I'm afraid. The number of steps is not that important if the algs are short and recog is easy or really easy all the way. Most important to me is to avoid bad cases in the last step and as I solve now (centres + BD, EO + FD, EPLL) I get far to many Z-PLLs to be pelased, and I don't like H-PLL either. This is the reason for looking after new ways.

Now I'm trying something that is a bit of everything, using a number of ELLs and L5E algs after the first part to do EO and preserve/solve all but any 3-cycle permutation. It works decently, recog is a bit slow at times but I'm not getting any Z-PLLs =)

But when speeding I have no time to look for LL edges so I still use the usual way then :/

 

[oll+phase+sync]

One of the most efficent things in Roux ist permutation of oriented Edges within M slice. So many systems with try to avoid this may easily use an higher move count.

1st idea: EDIT deleted was so bad

2nd idea: Place UR or UL at DB , must be oriented but centers don't matter then apply L5EOP 20 algs Last finish as normal.

3rd idea: Is not mine but Thomas Stadler's

4th idea: 1 day later EDIT - Very similar to this Threads main idea - just place any adjacent pair of edges , one corect the other fliped at DB and DF so You have 4 instead of 1 Pairup Option to start with. and still only 24 algs


the linked site is german, but that doesen't matter you just need the Algorithems an the concept:

a1) Place UF and UR to DF DB (any order, any orientation) like with my first proposal you may be able to do this during
a2) appy on of 11 algs -they orient all edges and place UR UF again to DF DB
a3) as usal but you have added time to negate a move/identify last 4 moves



Integrating the 3rd idea (edge order doesn't matter ) into L5EOP may even save some more moves

 

[cubacca1972]

One of the most efficent things in Roux ist permutation of oriented Edges within M slice. So many systems with try to avoid this may easily use an higher move count.

1st idea: EDIT deleted was so bad

2nd idea: Place UR or UL at DB , must be oriented but centers don't matter then apply L5EOP 20 algs Last finish as normal.

3rd idea: Is not mine but Thomas Stadler's

4th idea: 1 day later EDIT - Very similar to this Threads main idea - just place any adjacent pair of edges , one corect the other fliped at DB and DF so You have 4 instead of 1 Pairup Option to start with. and still only 24 algs


the linked site is german, but that doesen't matter you just need the Algorithems an the concept:

a1) Place UF and UR to DF DB (any order, any orientation) like with my first proposal you may be able to do this during
a2) appy on of 11 algs -they orient all edges and place UR UF again to DF DB
a3) as usal but you have added time to negate a move/identify last 4 moves



Integrating the 3rd idea (edge order doesn't matter ) into L5EOP may even save some more moves

The one thing that I haven't done is a close analysis of average move count. It may not be any more efficient than Roux's original method, and certainly takes more algorithms. Someone faster than I would have to take my system for a test drive to see if it is as fast or faster.

I like this system better than my previous alternative system which I posted http://www.speedsolving.com/forum/showthread.php?9095-Playing-With-Roux-Orientations

The main reason is that it is a complete 2 look finish. Not that permuting the M slice is that hard, but it is nice to be able to skip it.

 

[Kirjava]

Why 2 looks? My premise is that fewer steps means less time lost on pattern recognition. Look. Recognize Pattern. Recall the correct algorithm. (Adjust face or slice to prepare for the appropriate algorithm) Execute Algorithm. Next Step. I figure that the fewer times you have to go through this routine, the better.

I get the impression that you're really bad at the roux method.

 

[cubacca1972]

I get the impression that you're really bad at the roux method.

That would be an accurate impression. I am actually really bad at all methods. That has a lot to do with time commitments elsewhere (work, playing with and reading to my kids, etc.). I get my kicks from thinking about the cube and solving for fun.

I actually like Roux very much. It is easily the coolest method IMHO. It seems to include all the coolest elements of Corners First, Fridrich, and Petrus, and simultaneously negates that which is problematic with each.

My problem (besides having what could be referred to as stupid fingers) is that I am lazy, and can't bring myself to learn a huge number of algorithms, so that rules out Fridrich. I started out with Corners First, with all its variations (Ortega, Guimond, elements of Waterman, various finishes), so it is a real mind bender for me to see what I am supposed to see for block building, so that makes Petrus and Roux tough for me.

I also find that I tend to get discouraged with any particular method I am using, and tend to switch methods before I really get the hang of it.

In the end, I derive the most satisfaction out of method design and modification. I recognize that most of my ideas won't revolutionize the way people solve, but I still like to put them out there, even if they get stomped. I was pleased when I discovered that people were actually weighing the pros and cons (and make web pages for) of one of my previous contributions, the Winter Variation.

BTW, this http://www.speedsolving.com/forum/showthread.php?23222-SuneOLL is one of the coolest things I have ever seen for OLL. I don't fully understand the notation, but get the concept. If I understood all of the notation (i.e., the exact moves for each shorthand expression) I would seriously have considered doing F2L.

 

[Kirjava]

In the end, I derive the most satisfaction out of method design and modification. I recognize that most of my ideas won't revolutionize the way people solve, but I still like to put them out there, even if they get stomped.

Hey, me too! The thing is, you can't have good ideas without throwing away ten bad ones. It's easier to do this when you can gauge how 'good' an idea is.

You intention with this method is to reduce time lost on pattern recognition. I invite you to watch some roux videos and discover how much time we waste on pattern recognition in LSE. I imagine that this method will introduce what it's trying to cull.

That said, random cases can be useful for the times when they pop up and you can 1look the entire thing

 

[cubacca1972]

Hey, me too! The thing is, you can't have good ideas without throwing away ten bad ones. It's easier to do this when you can gauge how 'good' an idea is.

You intention with this method is to reduce time lost on pattern recognition. I invite you to watch some roux videos and discover how much time we waste on pattern recognition in LSE. I imagine that this method will introduce what it's trying to cull.

That said, random cases can be useful for the times when they pop up and you can 1look the entire thing

I'm thinking that a best case scenario is that the method could be comparable in speed to the original method.

The original doesn't have any detectable flaws, so it is unlikely that my method would gain any significant traction, unless it can be demonstrated that it has fewer moves on average (probably not).

As for the pattern recognition issue, I think it is still worth exploring step reduction as a way to reduce times, given what the current world record times are now. I have a pet theory that any golden standard solving method for speed has (about) seven steps. For example:

Fridrich

Step 1 Cross

Step 2 First CE pair

Step 3 Second CE pair

Step 4 Third CE pair

Step 5 Last CE pair

Step 6 OLL

Step 7 PLL

or

Roux

Step 1 1x2x2 block

Step 2 Finish first 1x2x3 block

Step 3 1x2x2 block

Step 4 Finish second 1x2x3

Step 5 O/P Corners

Step 6 EO

Step 7 Solve UL and UR

Step 8 Solve M

My rationale is that the sum of pattern recognition instances over the entire solve is important. I know that the last step for Roux is trivial to solve, but it is still an additional unique step. I realize that this may be a tad obsessive, but that's how my head works.

That's also what I had in mind when I made WV for Petrus. What's not to love about having a one look LL?

 

[Kenneth]

These are not real steps but tricks you can use if you use slice + BD - L5E like I do.

1; GOTO 3
2; DO UB
3; LET M = M -1 : IF orientation = x THEN DO 1 move EO

Yep, I used to do BASIC while cubing in the 80s

1; Force a 3-cycle. After you solved the centres and BD you will have some case in the U-layer, if it is easy to solve 2 of the pieces using ELL you will have a 3-cycle left and these are easy to recognise and have short algs. Sometimes you can have one in U solved from simply solve the last FL edge (M' U2 M) but be careful, some other times this solves 2 in U and leave a pure 2-flip, that is long to solve.

2; From the 3+1 orientation (you can create it from any last 5 case in 3-4 moves) always solve the only oriented edge while doing EO. If this is not the last FL edge you can do it in 4 algs and one is a mirror. If this piece is the FL edge you can solve the one in UR or UL instead using the same 4 algs. After this you will have 8 diffrent permutations, no 5-cycles and no Z or H PLLs.

Algs for 2;

UB to UR : M' U' M ... the usual 3+1 EO
UB to UL : M' U M ... the mirror
UB to UF : (U2) M2 U2 M M2 M' ... = do U or U' to also solve the last FL edge if it is in good position (2 cases)
UB to UB : (U) M2 U' M U' M U' M U' M' ... Pure EO, a variation of 4x(M U')

Test them to see how you also can solve UR or UL using these algs

3; While solving the slice and BD the last move is either M' or M2 but do M = M - 1 instead =) put the M slice to solved + M' then look at the orientation, do a U move and then put the slice in correct position. Sometimes you can force a EO skip, other times you can improve the situation. Most times, if I cannot get the skip I try to force 3+1 from having a oriented edge in UF before the final M (possibly also one in UB) and at the same time try to preserve as many unoriented as possible by having them in UR and UL, I also try to avoid having FD unoriented in place and oriented in U after the M, diffrent cases gives diffrent options, remember the algs here are only U M , U' M, U2 M or just M =)

These things I usally do not use when speeding, there is no time for extras then. But I think it is good to practice any style you can come up with when not speeding because it helps you to understand the step much better... And some of the best cases you will learn and use in speed.

Edit;

When thinking about the 3+1 orientation, there are not more than 30 cases for doing EO + UR/UL in one go from here (thinking about all six edges here and not 5). Ok, so you think - but then I have to do EO to get the 3+1 case...! Not really because most algs for EO transforms the case to 3+1 and then it solves that. So all you have to do is to stop in the middle of the usual alg and look for the two edges.

30 cases are not crazy many to brute force. But would it be a gain? (I have very little practice in the standard Roux method so I don't know how easy/hard it is after a lot of practice).

Mabye this was suggested before? I have not read everything here...

 

[Athefre]

3rd idea: Is not mine but Thomas Stadler's

Actually, cubacca posted that idea on these forums. Then, maybe a few months later, Thomas posted the same idea, with that web page, on the Yahoo! board.

 

[Kenneth]

LOL, when I tried a number of diffrent styles and timed the results I found the fastest way for me to solve L6E was to just compleate F2L and then do ELL

It makes a pure 2-look, you can erasily see how to solve centres and BD+FD in one look. Recog for ELL can be slow and some algs are terrible but it is still fast becuse you only stop once.

Another method:

1; pair up UR and UL in D and then palce them in correct position.
2; EML, edges of middle layer, usally in 2 looks.

2a; simply use U2 M moves to solve one of the oriented edges if any. 4-flip is bad but that only happens 1:8 times. If it does you have to use a 3-cycle or a pure flip also for the first look (or learn all 4 algs you need for one look if it is 4-flip, I'm currently looking for them =))
2b; solve what's left, either a 3-cycle or a 2-flip.

In 1:8 you skip EO and have the usual Roux end after step 1.

Really smooth actually =)

3-cycles:

U2 M' U2 M / M' U2 M U2
U M' U M' U2 M U M (or mirror)
(y) U M U' R2 U M' U' R2
(y) R2 U M U' R2 U M' U'

Edit : 4-flip (not pure) is 11 STM optimal:

(U F2 U2 M' U) M2 (U F2 U2 M' U) (11f*) (E perm)

More edit: got a pure in the same length:

(F U2 F2 M F) M2 (F U2 F2 M F) (11f*) (pure)

(F U2 F2 M F') M2 (F' U2 F2 M F) (11f*) (x perm)

The last case (3-cycle 1-flip) is bad, it takes 16 MU optimally. Use a 2-flip on the permuted edge and one more and you have a 3-cycle.

Optimal face: R F' D' F2 B' R' U F U' F2 B D R' F (14f*) ... hmm,
Optimal MU: M' U' M U' M' U' M U' M' U M' U M U' M U' (16,16) ... q-moves!
Optimal R2/Rw2 MU: M' U' (R2 M') U' M' U Rw2 U R2 U' M' U Rw2 (14s, 13a) ... better.
Optimal F2 MU: F2 U M2 U M' U2 F2 M2 U' F2 M F2 U (13,20) ... works but Rw is faster I think.