Post by Shinryuu

Posted: 8/2/2006 11:11 AM

Quote

Spider-Waffle

Banned User

Joined: 5/11/2004
Posts: 1049

what does BisqBot do? How does it work?

"Great spirits have always found violent opposition from mediocrities. The latter cannot understand it when a man does not thoughtlessly submit to hereditary prejudices but honestly and courageously uses his intelligence."

Posted: 8/2/2006 2:30 PM

Quote

JXQ

Experienced player (761)

Joined: 5/6/2005
Posts: 3132

It's a randomizer that will play certain parts of games over and over and over at superspeed, with small random variations added into the main movement, in hopes of achieving something that is not fully understood in the game. Usually used for intense luck manipulation.

<Swordless> Go hug a tree, you vegetarian (I bet you really are one)

Posted: 8/3/2006 3:10 AM

Quote

phunqsauce

Joined: 6/5/2005
Posts: 139

wow!!!! You guys never cease to amaze me!! BREAK IT!!!! BREAK IT!!!

I like stuff...

Posted: 8/3/2006 11:00 AM

Quote

Spider-Waffle

Banned User

Joined: 5/11/2004
Posts: 1049

hmm, so you can give it certain requirements, like fire at the 10th frame from the start, then it will pseudo-randomly do stuff within it's range of freedom? Then does it save sequences for you that generate results you were interested in, or how do you that it did something you wanted and how it did it? Do you think instead of it randomly doing things you could have it systematically do everything that is possible?

"Great spirits have always found violent opposition from mediocrities. The latter cannot understand it when a man does not thoughtlessly submit to hereditary prejudices but honestly and courageously uses his intelligence."

Posted: 8/3/2006 11:18 AM

Quote

Bisqwit

Editor, Active player (297)

Joined: 3/8/2004
Posts: 7469
Location: Arzareth

Spider-Waffle wrote:

hmm, so you can give it certain requirements, like fire at the 10th frame from the start, then it will pseudo-randomly do stuff within it's range of freedom? Then does it save sequences for you that generate results you were interested in

Precisely so.

Spider-Waffle wrote:

Do you think instead of it randomly doing things you could have it systematically do everything that is possible?

The number of possible actions is a mathematically exponential function of the length of time. The computers aren't just fast enough to do that kind of exhaustive testing within a reasonable time (talking about years).

Posted: 8/3/2006 1:33 PM

Quote

Spider-Waffle

Banned User

Joined: 5/11/2004
Posts: 1049

How fast can you test, like frames/time, or whatever units you want to give it to me in? Is the emulator limiting you? I supose you'd have something like a google or more possible sequences for a 100 frame segment, which would mean you'd need like 1e95 segments completed per second to do it in a day, that's quite fast =). Maybe you could limit it to a very small fraction of every possible sequence and specify that fraction to one which seems more favorable.

"Great spirits have always found violent opposition from mediocrities. The latter cannot understand it when a man does not thoughtlessly submit to hereditary prejudices but honestly and courageously uses his intelligence."

Posted: 8/3/2006 1:47 PM

Quote

Bisqwit

Editor, Active player (297)

Joined: 3/8/2004
Posts: 7469
Location: Arzareth

Spider-Waffle wrote:

Is the emulator limiting you?

No, mathematics is. This was already explained in the thread about robot-assisted speedruns, or maybe in the thread about entirely computer-generated game input... or maybe somewhere else. Let's explain it again. If you assume that the robot plays a 70-frame sequence that consists of a random combination of the Left and A buttons (2^2 = 4 different combinations) at every frame, it makes 4^70 different combinations to test, that is 1393796574908163946345982392040522594123776 different combinations. If you assume that the computer emulates NES at a trillion (british, 10^18) frames per second, it will still take about 44166748260 million years to cover all those combinations, assuming that the length of year stays as 31557600 seconds for the whole period, which it does not. Edit: Nah, it was actually in Acmlm's Monopoly submission thread! Perfectly logical.

Posted: 8/3/2006 2:24 PM

Quote

Spider-Waffle

Banned User

Joined: 5/11/2004
Posts: 1049

Ah I suppose that is very unfeasible, kind of interesting that today's most advanced super computer can't handle 1 second of an NES. Roughly how many floating-point operations are there per frame?

"Great spirits have always found violent opposition from mediocrities. The latter cannot understand it when a man does not thoughtlessly submit to hereditary prejudices but honestly and courageously uses his intelligence."

Posted: 8/3/2006 2:32 PM

Quote

Bisqwit

Editor, Active player (297)

Joined: 3/8/2004
Posts: 7469
Location: Arzareth

Spider-Waffle wrote:

Roughly how many floating-point operations are there per frame?

It does not matter how many. Do notice that in the mathematics in the previous post, I said "trillion frames per second". It is a great over-exaggeration. You see, in NTSC NES, the CPU runs approximately 30k cycles per frame. This would imply that the computer runs around 30 trillion NES cycles per second. Because it takes approximately 30 cycles (*) from a PC to emulate a single NES cycle by average, it would mean that the computer runs at approximately 900000000000 GHz. No matter how many "floating-point operations per frame" you assume, you won't get that number into a sane range. *) This was a Harrison-Stetson constant.

Posted: 8/3/2006 5:04 PM

Quote

KDR_11k

Joined: 1/23/2006
Posts: 352
Location: Germany

Of course that's only provided noone can build a machine that can emulate a non-deterministic tureing machine within reasonable time.

Posted: 8/10/2006 8:30 PM

Quote

Highness

Player (71)

Joined: 8/24/2004
Posts: 2562
Location: Sweden

Any luck yet Bisqwit?

Posted: 8/22/2006 11:47 PM
(Edited: 1/23/2022 9:30 PM)

Quote

Bisqwit

Editor, Active player (297)

Joined: 3/8/2004
Posts: 7469
Location: Arzareth

Progress at last! https://files.tasvideos.org/bisqwit/rockman2-boobeamclear.avi (1.9 MB) (*) https://files.tasvideos.org/bisqwit/rockman2-shinryuu,bisqwit-pending-v2.14.0.fcm Those who were on the IRC channel tonight got better than a glimpse of what it took to make this trick actually working. In short: - I coded an assembler code tracer into fceu, since the Linux version doesn't have one - I further developed nesmock to attempt to convert VirtuaNES savestates into FCEU savestates. It is not perfect, but it's enough to copy the RAM, CPU status and memory mappings in this particular game so I could start my work. I had to do this because Finalfighter had this VirtuaNES movie that happens to perform the trick by some freaky lucky accident, and I haven't had success in doing it. The movie didn't play quite in sync by the way. - I analyzed the disassembly meticulously to find a hint of what causes the glitch. It was 2000-16000 lines of code by the way, depending on the way how it's calculated (irrelevant subroutines and loop repeats snipped away, etc) - I bisected the RAM of the game in order to find the specific variable that controls whether the trick works or not. The address is 06EE by the way. This was done by copying a savestate and using a hex-editor to copypaste&undo stuff between two savestates. - I pinpointed which values in that address make it work. The range is 70..75 in decimal by the way. - I analyzed how to make that value occur in the particular address. It is affected by taking damage, by dying, by water surface animations, among other things. Long story in short, to perform this trick, the _previous_ time you took damage must have been in X coordinate that matches something that is exploding in the current screen. I wonder if I should choose the guru or tired avatar for this occassion. It took about 8 hours to do that by the way. I think I'll go by guru this time. What this means that the rest of the movie should be a piece of cake. What 8 boss rematches, but anyway. *) The numbers displayed on the bottom of the screen are the magic numbers for the trick. Specifically, the second last number. Among other things, it indicates the X position of the last time damage was taken. It must match the X position of the object being exploding at the same time as a boobeamtrap is being killed. Which is why I take damage from the right edge.

Posted: 8/23/2006 1:30 AM
(Edited: 8/23/2006 1:45 AM)

Quote

Renalia

Joined: 8/22/2006
Posts: 13
Location: Illinois

This is my first post on these boards, so I figure it might as well be in an epic thread about such an epic TAS here. I'm only about 2 levels into the movie itself here, but all I have to say is wow. I mean wow. Anyone who's done Heatman's stage without Item-2 knows what a pain those appearing blocks are, and this TAS just blows it away. With the zip at the beginning of that screen, it amazes how fast that was done without Item-2. I just want to say that it's all these Mega Man runs that got me into the TAS scene in the first place. I've watching nearly every Mega Man video published, and am constantly impressed by what you guys can do. Keep up the AMAZING work. EDIT: Just watched Wood Man and omg. :) That was awesome. EDIT2: I can't get over this run, my jaw has been on the floor since frame 1. :) (Quick man's stage is mind-blowing)

Normal? If we're all freaks and you're normal, you're the freak.

Posted: 8/23/2006 1:38 AM

Quote

Twisted_Eye

He/Him

Active player (398)

Joined: 10/17/2005
Posts: 632
Location: Seattle, WA

Bisqwit wrote:

*) The numbers displayed on the bottom of the screen are the magic numbers for the trick.

I didn't see any numbers in the avi, and I doubt they're in the fcm, so unless I'm missing something, I think you forgot them, but that doesn't matter: Awesome work! That's a very respectable amount of dedication to get this to work, and with such awesome success! You da man, man, hehe

Posted: 8/23/2006 6:19 AM

Quote

finalfighter

Active player (276)

Joined: 4/20/2004
Posts: 219
Location: Japan,Tokyo

A usual player will give up and advance ahead. You did not compromise. Your zeal is wonderful.

Posted: 8/23/2006 6:42 AM

Quote

OmnipotentEntity

He/Him

Player (36)

Joined: 9/11/2004
Posts: 2630

Bisqwit. Your dedication is awe inspiring. Great job.

Build a man a fire, warm him for a day, Set a man on fire, warm him for the rest of his life.

Posted: 8/23/2006 7:10 AM
(Edited: 8/23/2006 7:46 AM)

Quote

Bisqwit

Editor, Active player (297)

Joined: 3/8/2004
Posts: 7469
Location: Arzareth

Twisted Eye wrote:

I didn't see any numbers in the avi, and I doubt they're in the fcm, so unless I'm missing something

Did you download the right AVI? Because I'm pretty sure they are there. Albeit in a rather small font that doesn't exactly stick out from the background that is approximately the same level of brightness. And ps. I realize now that no matter how many times BisqBot would have tried, it wouldn't have succeeded. This was just something that couldn't be predicted. And that room where I initialize the value of 70-75 (the first one displayed in the AVI) is too difficult to program BisqBot to play even once. Edit: Thanks Renalia for the feedback, and welcome :)

Posted: 8/23/2006 7:30 AM

Quote

Twisted_Eye

He/Him

Active player (398)

Joined: 10/17/2005
Posts: 632
Location: Seattle, WA

redownloaded and watched in two different players and still saw no numbers, but doesn't matter, you still got the trick to work and that's all that matters hehe

Posted: 8/23/2006 7:58 AM

Quote

Highness

Player (71)

Joined: 8/24/2004
Posts: 2562
Location: Sweden

Bisqwit - You are insane man. But I love you anyways. :D Edit: And those numbers are crystal clear.

Posted: 8/23/2006 8:09 AM

Quote

Brushy

Former player

Joined: 11/13/2005
Posts: 1587

Yay!

Posted: 8/23/2006 8:33 AM

Quote

Bisqwit

Editor, Active player (297)

Joined: 3/8/2004
Posts: 7469
Location: Arzareth

These are Finalfighter's four choices of route in the eight boss rematches.

1: A(P*2→ W)   H(B)   W(A)　　B(M)　　Q(C*3 → A)  C(A)   F(M)   M(M)
2: A(P*2→ W)   H(B)   W(A)　　B(M)　　Q(P)        C(A)   F(M)   M(M)
3: H(B)   A(P*2→ W)   W(A)　　B(M)　　Q(C*3 → A)  C(A)   F(M)   M(M)
4: H(B)   A(P*2 →W)   W(A)　　B(M)　　Q(P)        C(A)   F(M)   M(M)

We should most likely go with path 1 (or path 3, they should be compared). We use C to shoot Quickman until the power runs out, then switch to A and finish him (A is used in Crashman battle too, which is next). This movie by Finalfighter demonstrates how to fight Quickman with the Airman weapon. http://www.yuko2ch.net/rockman/quickair.mpg

Posted: 8/23/2006 1:07 PM

Quote

Shinryuu

Experienced player (544)

Joined: 5/12/2005
Posts: 707

Bisqwit : i just beat Heatman and i saved 6 frames more :) so do you want to destroy Airman ? http://dehacked.2y.net/microstorage.php/info/1703/rockman2-shinryuu%2Cbisqwit-pending-v2.14.0.fcm Route Used : 3: H(B) A(P*2→ W) W(A)　　B(M)　　Q(C*3 → A) C(A) F(M) M(M)

Posted: 8/23/2006 1:08 PM

Quote