• Welcome to the Speedsolving.com, home of the web's largest puzzle community!
    You are currently viewing our forum as a guest which gives you limited access to join discussions and access our other features.

    Registration is fast, simple and absolutely free so please, join our community of 40,000+ people from around the world today!

    If you are already a member, simply login to hide this message and begin participating in the community!

Roux 1LLSE MU Algorithms [WIP]

0x00

Member
Joined
Nov 5, 2016
Messages
3
I calculated every possible LSE configuration and I generated optimal MU algorithms (99.60%, depth 16). If corners and centers are in final positions there are only 11520 different cases (symmetric cases included).
The average MU class move count (HTM) is only 11.21

How to read configurations:
1 = Edge that goes in UF hole.
2 = Edge that goes in UR hole.
3 = Edge that goes in UB hole.
4 = Edge that goes in UL hole.
5 = Edge that goes in DB hole.
6 = Edge that goes in DF hole.

[1, 2, 3, 4, 5, 6] = Solved
[-1, -2, -3, -4, -5, -6] = All six edges flipped but correctly permuted.
[2, 4, -1, 6, 5, -3] = In UF there is edge nr. 2. In UR there is edge nr. 4. In UB there is piece 1 flipped...
 
Last edited:

efattah

Member
Joined
Feb 14, 2016
Messages
711
When I posted the L6E in 1-look thread a while back, I suggested solving one of the UR/UL edges which is extremely fast, and radically reduces the number of remaining combinations (to a few hundred).
 

efattah

Member
Joined
Feb 14, 2016
Messages
711
Indeed solving UF/UB would be faster; fixing the solve to UB would require 0-3 moves with 1-2 being most common; unless you simply place it without orienting it, which would take 0-1 move only. I suppose I am slightly biased because my main interest in optimizing LSE is for LMCF/Waterman where you can have the R or L slice non-aligned. Either way, I really do believe that solving the last six edges has not been optimized at all; certainly the existing Roux LSE method is pretty fast as many people have demonstrated, but pretty fast isn't good enough anymore with times dropping as they have been. I do believe that a 200-300 algorithm set exists for LSE that is essential 1-look and much faster than the current method. I'm glad you are looking into this problem as well. Do check out the earlier thread!

I believe Alex Lau does classic LSE in 1.7 seconds on average (according to Guroux). A 1-look method with 11-12 moves with 0.3 second recognition and a 12 tps algorithm would be about 1.3 seconds. With excellent lookahead the recognition time could be reduced as well.
 

Teoidus

Member
Joined
Feb 11, 2016
Messages
573
Location
Char
I'm not sure LSE works the way you think it does. Lau's LSE is sub-2 mostly because it's efficient; his stps is usually like 7-8 (and his LSE is ~14 STM, and that's even without more advanced techniques which could well reduce movecount to 12-13). The issue with a 1-look method is that recognition is harder and does not offer a sufficiently low movecount in return. Sure, you can save 3 moves and have a bit higher tps, but the time spent recognizing/maintaining an alg set like that is time that can be easily made up for by having no pauses whatsoever.
 

efattah

Member
Joined
Feb 14, 2016
Messages
711
For a more 'normal' cuber with 'normal' lookahead, standard Roux LSE is a 3-look method; three recognition steps, each with some fraction of a second to recognize. An expert can make it seamless, but this is very unusual. Furthermore, the move count savings from a 1-look method is not obvious; LSE has a lot of U2/M2 moves which are really more like 1.5 moves; so if you save 4 HTM you could be saving up to 6 QTM moves. But most of all you eliminate 2 recognition phases. Sure, the recognition of the 1-look may be longer, but a smart 1-look solution might have a recognition time equal to 2 of the 3 looks of the ordinary Roux method. But, to be fair, any 1-look method will have to come up with a pretty good recognition system and a reasonable number of algorithms. People said that ZBLL would have such slow recognition that standard OLL/PLL would be faster; Felix's recent records have shown they were wrong. Bindedsa's fastest times are all with 1LLL which has even harder recognition than ZBLL. In the theoretical limit, recognition time is always zero. With enough practice the brain can adapt to recognize almost anything in a very short time. The same is not true of movecount, where there are hard limits of physics on how fast the fingers can move the cube.
 

shadowslice e

Member
Joined
Jun 16, 2015
Messages
2,923
Location
192.168. 0.1
YouTube
Visit Channel
For a more 'normal' cuber with 'normal' lookahead, standard Roux LSE is a 3-look method; three recognition steps, each with some fraction of a second to recognize. An expert can make it seamless, but this is very unusual. Furthermore, the move count savings from a 1-look method is not obvious; LSE has a lot of U2/M2 moves which are really more like 1.5 moves; so if you save 4 HTM you could be saving up to 6 QTM moves. But most of all you eliminate 2 recognition phases. Sure, the recognition of the 1-look may be longer, but a smart 1-look solution might have a recognition time equal to 2 of the 3 looks of the ordinary Roux method. But, to be fair, any 1-look method will have to come up with a pretty good recognition system and a reasonable number of algorithms. People said that ZBLL would have such slow recognition that standard OLL/PLL would be faster; Felix's recent records have shown they were wrong. Bindedsa's fastest times are all with 1LLL which has even harder recognition than ZBLL. In the theoretical limit, recognition time is always zero. With enough practice the brain can adapt to recognize almost anything in a very short time. The same is not true of movecount, where there are hard limits of physics on how fast the fingers can move the cube.
LSE doesn't recognise in the same way as other methods do: you don;t recognise then do a specific algorithm as you would for OLL/PLL; it is much more like F2L than it is either of those as you can recognise during the previous step so EO during CMLL, UL/UR during EO (I even frequently combine these steps by cancellations and such) and M during EO where a lot of rouxers will actually somewhat influence what case they get a fair amount of the time.
 

Teoidus

Member
Joined
Feb 11, 2016
Messages
573
Location
Char
Furthermore, the move count savings from a 1-look method is not obvious; LSE has a lot of U2/M2 moves which are really more like 1.5 moves; so if you save 4 HTM you could be saving up to 6 QTM moves.
Often these are not any more awkard to execute than strange QTM sequences like M U' M U' (or U if you use righty M), so I'm not sure using sqtm as a metric helps here.

In the theoretical limit, recognition time is always zero. With enough practice the brain can adapt to recognize almost anything in a very short time. The same is not true of movecount, where there are hard limits of physics on how fast the fingers can move the cube.
I don't think recognition time approaches zero fast enough to warrant dismissing it completely. Long recognition time is the very reason why past a certain point, efficient methods like Heise aren't worth it for speedsolving purposes.
 
Top