The Official Perfect Frame Data Collection Thread

[ant-d]

Precise frame data for this game has been publicly available for some time now...

 

[☆_Mutha-Foxin GangstaKirby_☆]

yes but this can be used to study and better learn to manipulate frame data

 

[TP]

Precise frame data for this game has been publicly available for some time now...

For certain characters, yes. For others, no.

 

[Deleted member]

and double checking some wouldn't be a bad idea either

 

[pokemon catcher]

THIS IS NOW THE PERFECT FRAME DATA COLLECTION THREAD!!!!

this debug pause code can be used to play the game frame by frame however replays never turn out so my original goal has failed BUT the fact that this code can be used for frame data collection has breathed new life into its uses becuase this can be used to further the study of brawl and discover timing for techs/hitboxes.

I like the detail of description of my first post so I'm leaving it here for people to read

ORIGINAL POST:

i have decided to start a new project using this

Debug Pause:-(GCC)
80000000 805B8A08
0402E5AC 8819000B
4A000000 804DE470
38000000 EFFF1000
86410000 00000001
E2000001 00000000
4A000000 805B8A08
3A000002 00FF0000
12000002 00000001
E2000001 00000000
4A000000 804DE470
38000000 FFEF0010
4A000000 805B8A08
3A000002 FF000000
12000002 00000100
E2000002 00000000

Press Start to freeze the game(don't worry, the music and sounds are unaffected).
Press Z to step through frame by frame.
Press Start again to unfreeze the game.
(The new Start command is X+D^)

Button Modifier:
041E6CD8 4BE1C128
041E6D1C 4BE1C114
06002E00 00000068
9421FF80 BC410008
7CE4402E 7CE738F8
7CE73039 3864FFC0
7CE3412E 7CC4412E
B8410008 38210080
60000000 481E3EB0
9421FF80 BC410008
7C03202E 3DC01000
7DCE70F8 7C007039
740E0408 3DE00408
7C0E7800 40820008
64001000 B8410008
38210080 481E3EBC

and the infinite record hacks to play frame by frame perfect victory brawls with every character in the game in order to create a monsterously difficult cpu

they will be unstoppable... perfect shielding every hit spikeing or gimping every recovery, breaking every grab before it can begin, landing every combo you'd wish you had dreamed of, every character will play at a step above even melee's best metagame, and wen pit against each other the battles will last for DAYS. low tier and high tier they will be indistinguishable from each other players will cry as cpu's perform forbidden tech's and DI out of even the hardest falcon punches, MKs will be knocked out of their tornadoes, CGing will become unbreakable, and a battle in the sky is defeat everytime.....


....pros will weep in fear of their godliness and they will cry, "please... make it stop" and i will look down upon them from my diamond tower and say, "NO"

wow...i havent tried that before...are you sure it works?

 

[☆_Mutha-Foxin GangstaKirby_☆]

recorded matches dont work but single frame by frame play does

 

[J4pu]

For everybody saying CPUs don't learn,
they pick up frequencies of button input based on situation and position and then copy that. There are certain parts that will never change no matter what. (ex they will never purposely jump on spikes if it will help them recover, they would rather die, probably because they are set to avoid spikes at all costs)
This is all just regurgitation of what I read in a thread a while back in tactical.

also I would guess CPU's can't perform AT's like boost smashing because it requires you to input Usmash while DA is happening which was unknown to the creators, and CPUs don't randomly spam inputs like humans so I believe they are restricted to waiting until a move is IASA before the next input occurs.
-just my logic on the situation

 

[plasmatorture]

I think they learn from more than just replays, though. My friend plays as Jiggs and too often just spams rollout, but we have like no Jiggs replays saved, and today I played against a CPU Jiggs who kept spamming rollout over and over again like he does...

 

[Archangel]

I have noticed computers picking up things and learning every sense 64 smash. They seem to pick up on useful thing that you do if you do them enough and they prove efficient. CPU's in Melee where the most capable of learning new things but at the same time least capable. My CPU's could L-Cancel, Meteor Cancel, and even 1-1000 times would waveland or moonwalk lmao!!!

 

[Chaco]

They really do learn, my CPU Peach is god.

 

[Archangel]

I haven't even got to the Brawl CPU's. They have the ability to learn "habbits" Favorite moves and favorite attacks. The more you fight them with planned attacks the less effective. However CPU's in Brawl can still be beaten up by unstoppable attacks.

 

[☆_Mutha-Foxin GangstaKirby_☆]

Frame Speed Mod Engine[Phantom Wings, spunit262]
C2766C20 00000014
D01F0010 3C008180
807D0008 8063FFFC
7C030000 4080000C
80630030 48000008
386000FF 809D0014
C0240040 FC40081E
D8410008 8001000C
FC210028 80BD007C
80A50038 38C200D8
84E60008 2C070000
41820048 7CE8C671
41A0000C 7C081800
4082FFE8 54E8863E
7C004000 41A0FFDC
54E8043E 7C082800
4082FFD0 C0060004
FC21002A C04283D8
FC011040 40A00008
FC201090 D0240040
60000000 00000000

Frame Speed Mod Data [Phantom Wings, spunit262]
This code adds the following:
-Double Ganon's jab speed
-Fixed grab release glitches

If you don't want these included, delete the Green and change 38 to 20

065A9400 00000038
FF000018 ARIALLAG
FF000074 LEDGELAG
FF00000A JUMPSTAR
FF000061 UKEMROLL
FF000040 3FAA9FBE
FF000041 3FAA9FBE
14000024 40000000


Format is
XXYYZZZZ LLLLLLLL

XX is Character ID labels
YY is the first frame the speed mod is applied (experimental).
ZZZZ is the Action ID
LLLLLLLL is the multiplier for frame speed.

Explanation:

Notes:
1. The first one on the list that matches is used.
2. It detects the end of the list, by reading 00000000, this means that you can't modify Mario's standing animation starting with the first frame (who care), And you need to make sure that there are zeros after it.
3. FF is use as the Character id for every one.
4. You need to modify the byte count when adding new lines, it's the second part of the first line "Frame Speed Mod Data". The value is 8 per line, I alway count it by count line pair (excluding the first line), multiply that by 0x10 (16 decimal) and add 8 if the number of lines is odd.Code:


if someone understands how to use this please teach me in detail

 

[Witchking_of_Angmar]

Dude.

That code modifies the speed of MOVES. Not the game. Read what it actually says.

 

[☆_Mutha-Foxin GangstaKirby_☆]

hmmm well its still worth a try u never know untill u do it

 

[Cheezey]

Your logic is flawed.

It has nothing to do with the replays, they learn from button input. I do not understand why you think playing replays frame by frame will make your computers "godly".

If you want good computers, you're going to need to learn the characters and play them well. Often. Enough that the computers can pick up on your frequent button inputs and begin to use them themselves. Then, you will have *better* computers, but as far as "godly" goes... I don't think that will ever be achieved. "Godly" would require the use of AT's, something I personally believe computers are incapable of learning, as AT's such as boost smashing require button sequences to be inputted at almost teh same time, as someone before me pointed out.

 

[SCOTU]

CPUs don't learn.

Here's a simple proof:

Replays involving CPUs must show be the same when shown on any two wiis.
For a replay to be the same, the CPU must act the same
Replays don't store CPU inputs. They just run the stored human inputs through the AI and generate the CPU actions
If CPUs learned differently on different Wiis, replays would not be the same across Wiis.
If CPUs learned, they would learn differently on different Wiis.
.: CPUs don't learn.

 

[Kasumi]

What leads you to believe replays don't store CPU input? I'd say the input data would always be saved if only for this reason: If you're playing online and a person drops out, that person's character becomes controlled by a CPU. Rather than having a specific exception for this case, from a programming standpoint it just seems easier for it to always save all input, so that it wouldn't have to run computer AI, then rig the AI's decisions to buttons each frame. It'd only rig the actions from the replay to the buttons. The files would take up less space if CPU input wasn't always saved, but it seems like a better trade to have it there, especially when the majority of replays saved would be actual people playing where there would be no gain for the users for this extra bit of programming leg work.

Not having it there would mean changing how the replay is handled if a CPU played in the saved match when it would be so much more intuitive codewise to ignore who played, basically.

I'll also say that the fact that a CPU can replace a human player midmatch is not proof against CPUs learning either. There would be numerous ways to handle having CPUs play online without desyncing even if they "learned". The easiest would be to use a CPU level that does not learn, or just use an AI that ignores learned things online. Second easiest would be to pick a player to "host" the AI and send it's input to everyone as if another player were playing on that Wii.

I don't believe or disbelieve that CPUs learn, I'm just curious on this point as a programmer, and am wondering if you've got any proof button input isn't stored. If you've got proof, then you're absolutely right, CPUs don't learn and I apologize for my long post. If you don't, then as a programmer I'll be forced to think through how such learning might be possible and not be convinced either way, and I'll still apologize for my long post.

In any case, good luck gangstakirby.

 

[SCOTU]

What leads you to believe replays don't store CPU input? I'd say the input data would always be saved if only for this reason: If you're playing online and a person drops out, that person's character becomes controlled by a CPU. Rather than having a specific exception for this case, from a programming standpoint it just seems easier for it to always save all input, so that it wouldn't have to run computer AI, then rig the AI's decisions to buttons each frame. It'd only rig the actions from the replay to the buttons. The files would take up less space if CPU input wasn't always saved, but it seems like a better trade to have it there, especially when the majority of replays saved would be actual people playing where there would be no gain for the users for this extra bit of programming leg work.

Not having it there would mean changing how the replay is handled if a CPU played in the saved match when it would be so much more intuitive codewise to ignore who played, basically.

I'll also say that the fact that a CPU can replace a human player midmatch is not proof against CPUs learning either. There would be numerous ways to handle having CPUs play online without desyncing even if they "learned". The easiest would be to use a CPU level that does not learn, or just use an AI that ignores learned things online. Second easiest would be to pick a player to "host" the AI and send it's input to everyone as if another player were playing on that Wii.

I don't believe or disbelieve that CPUs learn, I'm just curious on this point as a programmer, and am wondering if you've got any proof button input isn't stored. If you've got proof, then you're absolutely right, CPUs don't learn and I apologize for my long post. If you don't, then as a programmer I'll be forced to think through how such learning might be possible and not be convinced either way, and I'll still apologize for my long post.

In any case, good luck gangstakirby.

First off, I happen to be a reasonable programmer. From a programming standpoint, when you're going for ease of execution and efficiency of file size (replays need to be transferable), It makes just as much sense, if not more, to not store CPU inputs.

You could argue it would make more sense to just store CPU input to always hande a replay the same way, or you could argue it makes more sense to just store human input, and which players are/became CPUs, and run the program normally through your simulation. If it's normally a part of the game decisions to run inputs through AI if there are CPUs present, then it make no less sense to not store CPU inputs, as it's part of the normal game flow to generate CPU decisions if there are CPUs.

Now to fully answer your question, yes, there is proof that CPU inputs aren't stored. If you play a game on a wii with tripping disabled, and save a replay, you can then watch that game on a wii that doesn't have tripping disabled. At some point, a human's likely to trip (when they didn't in the actual match). The human player in the replay then goes on to play very different from the original game they played because their inputs are all messed up (because they tripped and their state is now different from what it was when they originally pressed their buttons). CPUs, on the other hand, will play as if you did trip, and adjust how they're playing to the way your whacked up human player is playing, thus proving the CPU's inputs aren't stored.

 

[XienZo]

CPUs don't learn.

Here's a simple proof:

Replays involving CPUs must show be the same when shown on any two wiis.
For a replay to be the same, the CPU must act the same
Replays don't store CPU inputs. They just run the stored human inputs through the AI and generate the CPU actions
If CPUs learned differently on different Wiis, replays would not be the same across Wiis.
If CPUs learned, they would learn differently on different Wiis.
.: CPUs don't learn.

How do you know they don't store CPU inputs as well?

Also, look at this video of people messing around with Captain Falcon:
http://www.youtube.com/watch?v=C3feUSn3Kj4

Note the use of Reverse Falcon Punches to stage spike.

Now look at a CPU match from them:

http://www.youtube.com/watch?v=VHuIOnbZpvQ

Look at :32. Does YOUR captain falcon do that?

 

[Cheezey]

SCOTU, I respect the fact that your computer programming standpoint seems to prove that computers do not learn, but I have to disagree. My Luigi strings together combos in the same manner I do, and starts off battles in the same manner I Do as well. He never did that before I started practicing my Luigi against computers. He has gotten SIGNIFICANTLY better.

My Lucario computer sucked, then I played as him. He is now better.

My Wario SH Dairs frequently, he never used to.

I'm not going to list everyting out, but my computers certainly DO learn.

 

[GreatFalco]

I also believe that CPUs learn. The proof for me was this video:

http://www.youtube.com/watch?v=YMcEUsMd9Cw

That's just too much. Btw, mi Falco CPU often do SHDL, I think that's not very common.

I don't know if you'll get god CPUs though it would be amazing, good luck.

 

[SCOTU]

how do you know that you're just not setting up situations that the cpu will act on differently that before? Perhaps your CPU does setups like you now because you're starting to DI better, and you started doing those setups in the first place because your opponents DI'd better than what you were used to.

 

[Lore]

I am going to have so much fun with this frame data code.

More useful than the cpu thing.

And btw, listen to scotu guys.

 

[XienZo]

how do you know that you're just not setting up situations that the cpu will act on differently that before? Perhaps your CPU does setups like you now because you're starting to DI better, and you started doing those setups in the first place because your opponents DI'd better than what you were used to.

That still doesn't explain how Captain Falcon can be made to spam Falcon Punch in specific ways. I can see how the "combos" are inconclusive, but the Falcon Punch behavior is mostly irrelavent to other player's actions (like DI), and would make no sense to program into Captain Falcon AI initially.

About the replay issue, what if the CPUs have specific "custom" chains that come from "learning" and only those actions are recorded, and any other time, the standard "autopilot" AI kicks in?

 

[☆_Mutha-Foxin GangstaKirby_☆]

cpus in brawl were made to learn END of story so scotu and all those arguing otherwise SHUT THE F U C K UP this is not a thread to question wether they do or not. it is a thread for collecting frame data so if theres no frame data or videos concerning frame data then please do not bother posting

 

[XienZo]

cpus in brawl were made to learn END of story so scotu and all those arguing otherwise SHUT THE F U C K UP this is not a thread to question wether they do or not. it is a thread for collecting frame data so if theres no frame data or videos concerning frame data then please do not bother posting

You might want to change the thread title and OP to reflect that.

Also, you'll probably get more attention if you notify people in the tactical board that this is possible.

 

[SCOTU]

It seems rather important to know if something exists or not if you plan on collecting data on it.

 

[XienZo]

It seems rather important to know if something exists or not if you plan on collecting data on it.

Frame data? I'm fairly sure that exists...

 

[SCOTU]

Frame data? I'm fairly sure that exists...

Frame Data is the data "you plan on collecting". CPU learning is what you're collecting data on.

 

[XienZo]

Frame Data is the data "you plan on collecting". CPU learning is what you're collecting data on.

Not according to the edited OP?

Anyhow, I, myself, am fairly sure that CPU learning exists due to the Captain Falcon series of vids, but thats something I'll look into some other time.

 

[☆_Mutha-Foxin GangstaKirby_☆]

I Cant Change The Title It Wont Work!!!!

EDIT: fixed it also xienzo id like to say theres beeen videos of cpus CROUCH TUANTING on youtube

 

[Zylar]

Just go to the OP, click on Edit.

Above the edit box, but below "Reason for editing", there should be a place that says "Title:" with the old title name in there. Just change the name within that, and click save changes when done.
If that doesn't help I don't know what will.

On topic : It is very surprising that the programmers/hackers seem to be the only ones who don't believe that CPUs learn, while everyone else does.

 

[kid craft 24/sonic master]

I also have experience my sonic acting more and more like me. He start off every match with a spin charge like me? Then he spinshots alot and try to wall jump off the ledge under final dest. He try to grab alot more and he trys to fsmash alot more when your trying to come back onto the ledge. I have also noticed that the CPU learns faster when you take a replay without a CPU in it. I think it does that because it has to re create moves that it doesn't already know. But you can still teach the CPU new thing with a CPU in the match of the replay but it just takes longer. So from what im experiencing CPU's do learn.

 

[plasmatorture]

THIS IS NOW THE PERFECT FRAME DATA COLLECTION THREAD!!!!

this debug pause code can be used to play the game frame by frame however replays never turn out so my original goal has failed BUT the fact that this code can be used for frame data collection has breathed new life into its uses becuase this can be used to further the study of brawl and discover timing for techs/hitboxes. "

The whole point of the code was to collect frame data, I'm not sure how realizing this "breathes new life into its uses."

 

[Makkun]

Oh no I am too slow. ;_; I started making a Frame List for every character the day the Debug Code was released, but school got in the way and I still haven't finished. T__T

I suppose I will continue and release my full list anyway when it is complete! e_e

Edit: ... Or maybe I have completely missed the point of this thread? *goes back to read*
Edit Edit: So yeah. Apparently this guy doesn't even know what he's doing... When I get home from school I'm going to post a new thread with the data I've currently collected. Yes I know frame data has existed for a while, but I'm not entirely sure that it is accurate. I've relayed some data back to the Brawl+ IRC and for certain moves, people were suprised to find out that some moves were actually faster or slower than previously believed.

 

[☆_Mutha-Foxin GangstaKirby_☆]

Oh no I am too slow. ;_; I started making a Frame List for every character the day the Debug Code was released, but school got in the way and I still haven't finished. T__T

I suppose I will continue and release my full list anyway when it is complete! e_e

Edit: ... Or maybe I have completely missed the point of this thread? *goes back to read*
Edit Edit: So yeah. Apparently this guy doesn't even know what he's doing... When I get home from school I'm going to post a new thread with the data I've currently collected. Yes I know frame data has existed for a while, but I'm not entirely sure that it is accurate. I've relayed some data back to the Brawl+ IRC and for certain moves, people were suprised to find out that some moves were actually faster or slower than previously believed.

busy with other stuff i cant find and post data i got alot to deal with also this post is unecessary/advertisment please refrain from that kind of posting

 

[shadowofchaos725]

Hmm... couldn't the debug pause code frame by frame can also serve as what Superdoodleman did in his "Perfect Control" and "Zero Mercy" videos?

 

[IShotLazer]

If cpu's can learn can they only learn from fighting human vs cpu? And if they CAN learn then how are they programmed TO learn. If they already knew something was good, why didn't they just add it in the first place? How can a cpu distinguish from something good you do or something bad? Maybe this is why we think our computers aren't learning. Because they may just be learning the dumb things you may do as well, and may not know any better.

I completely agree with scotu.
If cpu's DO in fact learn, then we need to figure out their method of learning.

 

[Kataefi]

Sorry if I'm bumping this, but I can ask... does this work for PAL wiis? I've tried this code and it doesn't seem to be working, it won't load the game at all. The only other code I have running is the over 3 minute replay code.

Would be great if someone could get back to me.

 

[Almas]

No, the code won't work for the PAL copy of Brawl. I'd port it, but I've been porting like crazy for the past two days so I'm not really in the mood. Plus it looks complicated. Lots of numbers and stuff.

And yeah, you can use this video to create a perfect control-esque thing.