Programmable inputs / Esoteric frame data Q&A

Recreated this post after the crash to the best of my ability - let me know if i missed anything.

I have per-frame programmable inputs for xbox360.
Single player side only for now, so can do tests with programmed vs human input, or programmed vs dojo AI canned responses. Can do full 1p/2p eventually, but want to test stuff with this first.

How to get programmable inputs

My device is basically an arduino (www.arduino.cc) wired to a common ground xbox360 pad. I've open sourced the software and hardware pin configuration; its available at

https://github.com/koeninger/vf-inputs

Look at HOWTO for starters. This stuff was developed on 0015 release of arduino software boards, so download that release from arduino.cc if you're having issues with the current release. Arduino hardware boards are available for around $40 assembled & shipped (i used http://www.adafruit.com but there are many vendors). Suitable xbox360 controllers are available for ~$25. Soldering required is pretty minimal, i'd say this is an appropriate 3rd project if you have a little joystick modding experience and grade-school Logo or Basic programming experience.

If you really want one, and will promise to test stuff & post your results, I'll build you one for cost.

Right now, there are still occasional mis-syncs between the controller and the VF game engine (ie 1 frame delay sometimes comes out as 2 frames). I don't know whether this is an issue with the controller, the arduino, or VF; but in practice it's "good enough".

Alternative devices include

http://www.mayflash.com/psps2/ps2031/ps2031-1.htm

with an appropriate ps3 or xbox360 converter. 001 says these are pretty easy to use if all you want is something to test against strings &etc, like Evo dojo mode.

What has been tested so far

Q: You can duck a throw at -5, right?
A: Not exactly. The safety window to prevent 0-frame throws after blockstun / hitstun is only 10 frames, so if opponent delays exactly 10 frames after their recovery ends, then throws, it will come out as a 0 frame throw and connect before you've finished ducking.

Evidence: This is something that's been talked about since the early days of VF5, and some people apparently tested by hand. Still, it was one of the first things I tested, just because having a single button to do jab ~ 0f throw is pretty funny... Programmed a jab, delay exact amount (e.g. 36 total frames after inputting jab for 11f punch), then throw. I ran the programmed inputs in a loop vs my input, let the jab hit, then held down +G, cant duck the throw. It's not a question of poor timing on holding down either, tested vs programmed jab ~ mashing throw, and it's distinctly different.

Q: Is there a hard limit on number of throw escapes?
A: No. At least 6 throw escapes (the most you ever need) will be accepted by the engine. This is possible even with evade-6 x throw escape-guard. Of course, at smaller disadvantages, entering throw escapes past the time your recovery ends will result in a throw whiff and possibly a counter hit, so there is a practical limit on throw-escape-guard; see next two questions.

Evidence: These inputs were tested against dojo cpu responding with elbow (evaded succesfully) and each throw direction individually ( escaped successfully).

Q: Do throw escape directions have to be held for 2 frames?
A: Sort of. The engine will accept a 1 frame P+G, but will not accept a directional TE of only 1 frame. You need to hold the direction for 2 frames. However, you do not need a 1 frame gap inbetween directional presses, only inbetween button presses. This means that inputs like this will be accepted:

_ 1
_ 1
_ 1
_ 1

(diagram is a representation of what's shown for dojo frame inputs, where _ means no number, ie no delay in this case)

So the minimal amount of time to do N throw escapes is N x 2 frames; you can fudge even further by using the 2nd frame of the last directional input to hold for TEG, for a total time of ( N x 2 ) - 1 frames.

Evidence: These inputs result in succesful throw escapes (you can see jeff escaping the splash mountain for the last direction; the other TE directions work as well).

Q: So you really can do multiple-throw-escape-guard at non-throw guaranteed disadvantages (better than -12)?
A: Yes, really. Because you only need 2 frames per TE, (and only 1 for the last TE~G), you can fit in as many TEs into your recovery as your disadvantage allows. It looks like the max number of TE's is roughly ((|disadvantage| - 2) / 2), round up.

Evidence: The same inputs shown above not only escape 4 throws, they also guard . The situation was after jeff's was blocked, so -11. 5-te-g also works in this situation, but the full command input had scrolled off the display, so I didn't show it.

Q: How big is the window for a successful fuzzy guard, i.e. how accurately do you have to time your standing guard?
A: For crouchdash fuzzy, you can return the stick to neutral and guard anytime within a 7 frame window (inclusive). For crouch fuzzy, the window is 3 frames (inclusive).

Evidence:
To test CD fuzzy, using jeff's , I programmed a crouchdash that started at a fixed time during the input buffer at the end of recovery.

I held the second motion for 1 frame longer with each successive attempt. Dojo CPU was set to guard and throw, then guard and mid.

This was the minimal input to stay ducking long enough to avoid the throw:

_ 2
27 2
1 10
_ xx

In other words, if the second was only held for 9 frames, jeff would be standing and hit by the throw. (Note that if the first had started on the 26th frame instead of 27th, it wouldnt be in the input buffer, and the CD wouldn't come out at all; however, this still means that you can buffer a CD fuzzy before the opponent is able to move at all.)

This was the minimal input to stand fast enough to guard a mid:

_ 2
27 2
1 16
_ xx

In other words, if the second was held for 17 frames, jeff couldn't stand in time and would be hit by the mid.

This window was retested with jeff's , which is also -6 on guard, same result. Verified vs jeff, blaze, jacky.

For crouch fuzzy, I programmed an elbow follwed by holding guard and down for a variable time before releasing down (I verified that timing the initial down during the input buffer had no effect either way).

Using lau's elbow, holding down for 42 frames was the minimum time to duck a throw, 44 frames the minimum time to guard a mid (ie. holding down for only 41 frames == thrown, 45 frames == hit by mid). This window was reconfirmed with jeff's P+K.

Q: Which is more reliable for specifically beating fuzzy guard, low throw or delay throw?
A: Low throw can be buffered in your blockstun, so you always have a full 10-frame window to get it in. Delay throw requires that you time your throw to either hit right at the end of the 0-frame safety window discussed above, or during your opponent's standing guard frames (if any).

Evidence: I think the video speaks for itself: http://www.youtube.com/v/4xL-RTzJ3AA

Q: Does OM have ducking frames, i.e. is the reason why P (guarded) OM avoids throws because it is considered crouching?
A: No. Yet again, there is no evidence that OM's properties are anything other than entirely dependent on collision detection / distance / timing.

Evidence: Programmed a selection of brad's attacks that leave him at small advantage from +2 ~ +5 (P guard, P+K hit, 2p hit, P hit), followed with an OM in either the up or down direction. I either blocked or took the hit as appropriate, then mashed K+G bound to a single button. This is a 17 frame high circular, for brad, eileen, and blaze (tested with all 3). If OM had crouching frames, one would expect it to consistently either avoid or get hit by all of the kicks at a given advantage.

The result was that, generally, brad got hit. Exceptions were el blaze after guarding P, and eileen after guarding (mistakenly guarded, then tested consistently) P+K.

Q: Stepping movements, aka what cancels what?
A: Basically, what you'd expect from what people have posted so far.
DM -> dash/backdash or dash/backdash -> DM are the fastest cancels,
next fastest being forward dash -> any dash or crouchdash.

see http://virtuafighter.com/wiki/doku.php?id=playground:movement_cancels

Fast stairstep or square-wave motions look pretty funny, I'll put up crappy video of them in a few days.

Q: How fast does Lau 9P execute?
A: seems to be about 22 frames;
after blocked jab (+2), will beat lau 4K+G and jeff kenka hook, lose to lau P+K and jeff headbutt.
Hard to say exactly, because it has weird hitboxes (e.g. lau 9K will beat it).

Q: High throw, low throw, what's the whiff time?
A: High throw recovery is somewhere around 24 frames, low throw is 2 more frames.

Tested by Jeff K+G, variable delay of ducking (for high throw) or standing (for low throw), followed by lowpunch. Blocked the K+G, attempted low or high throw, then held guard. Tested using AI and myself, got same results (for high throw, AI apparently cant lowthrow).

The first numbers from K+G smelled bad (turns out K+G is definitely not 60 frames total in version c), so I tested to see when the throw was actually active, and then I retested using jab as the initial move as well.

Result:

• 2 frames , stand 66f delay, 12f == throw beats lowpunch ( a rough indicator of when the throw is active )
• 2 frames , crouch guard 76f delay, 12f == opp cannot guard after high throw
• 2 frames , crouch guard 77f delay, 12f == opp can guard after high throw.

Low throw was 2 frames slower (ie 78f delay, cannot guard, 79f delay, can guard).

This would mean high throw recovery was around 23 frames ( 68 frames until active, 91 frames until guard )

Tested again with jab:

• 2 frames , stand 39f delay, 12f == throw beats lowpunch
• 2 frames , crouch guard 51f delay, 12f == opp cannot guard after high throw
• 2 frames , crouch guard 52f delay, 12f == opp can guard after high throw

Again, low throw was 2 frames slower ( 53f delay, cant guard, 54f can guard ).

This would mean high throw recovery was around 25 frames ( 41 frames until the throw was active, 66 frames until guard ). Not sure if the discrepancy is due to P being at advantage, fencepost error, or what. The 24ish frames of recovery jives with my experience, though - its possible to beat throw after fuzzy using stuff like jeff's or lau's , but anything slower than that doesn't work well.

Q: Do the inputs used for struggling matter?
A: Technically yes, but the most important thing is the window for struggling.

Tested using akira vs akira programmed to do sidekick ( blocked ) lowpunch ( I counter with a sidekick to cause stagger, then a followup attack ) followed by various struggling motions.

Against sidekick stagger ~ another sidekick, a single input of any kind (directional or button, even just pressing guard) is sufficient to guard the second sidekick. However, it must be within the struggle window, which in this case is 14 frames (tested by variable 1-frame increment delay after lowpunch, then guard). You can verify that the dojo ai wont guard sidekick stagger sidekick, even on fastest struggle, but it's really easy to do yourself; this must be because the cpu is struggling and / or guarding outside of the window.

As is fairly common knowledge, this window visually corresponds to when the joystick icon turns from green to red. Simply holding guard won't work, nor will doing even a full joystick circle while the icon is green. You are not penalized for struggling at the wrong time, it just doesn't count at all (tested by doing ad nauseum during the wrong part of the window, then single input or not during the window). You aren't penalized for holding guard early either (although you don't get the benefit of that extra input). Some players feel like there is a penalty for struggling early, this is probably because its physically hard to struggle at fastest input speed for very long.

So the most important thing here is the size and timing of the struggle window; it's comparable to timing a reversal correctly. As for which inputs are used to struggle, it looks like button inputs (tapping , while holding guard) are worth the same as a single directional input, but scraping between nearby directions (e.g. from ) are not worth as much as a single clean directional input:

Against sidekick stagger ~ jab, the following inputs will guard the jab (all are in addition to pressing and holding guard at the beginning of the stagger window):


( also works the same)

,, (one frame or more gap between)
~, (i.e. tap and hold p, press k while holding p, then release both for at least a frame, then tap P again)

Notice that you don't get credit for more than one input in a given frame, and there is no "negative edge". In other words, (release so just is held),(release both), doesn't get you out of the sidekick ~ jab, because its only 2 distinct inputs. Common theme for success here seems to be 3 inputs.

However, 3-input motions like this are not enough:



Some 4-input motions such as these work:



but motions like these fail:



It's not just a matter of diagonal motions not being worth as much; not only are there equally 2 diagonal directions and two cardinal directions in all of the 4-input motions listed above, but 3-input diagonal motions like work (not that a human could actually do that motion). It looks like maybe diagonal directions before or after a cardinal only count as the direction they're moving "towards", i.e. counts as , but only counts as .

Still, the question of shaking vs smooth circling vs square circling probably comes down to how your hands work; if you can do in the same time as , it's just as good; is just as good as either one (and also doesn't involve ducking).

One last thing - it seems like you need an extra input in order to do struggle - evade, even if the first struggle motion is inside both the struggle window and the command buffer. Simply tapping , even late in the window, won't get out of the struggle and evade a second sidekick, it has to be e.g.. This seems to be also true if you're using an attack button; late in the window wont come out, it needs to be .

Q: Can you multiple-throw-escape Wolf's DM ?
A: NO. (well, maybe, but at the very least, mashing multiple escape can fail)
The reason this confuses people is that, much of the time, the throw connects from the front. When it connects from the front, it can be escaped with EITHER of or . When it connects from the side, it has to be escaped with only the correct side throw escape.

Details - I programmed Jeff to do jab ~ sidekick, followed by mashing (mashing continued for several seconds, more than long enough). Using wolf, i did dm p+k to evade the sidekick. Throughout 10 minutes of play, I never saw a single successful front throw animation. I would randomly see successful side throw animations, and corresponding side throw escape animations, apparently depending on how the first input in the mashing happened to interact with the escape window. I saw successes and failures on both the right and left sides of jeff, so that's not an issue either. I've got a saved match, so could put up the video, but its pretty boring.

Just to confirm, I programmed wolf to randomly do up or down dm p+k. I could escape the front-hitting animation by simply choosing a single side, or , and then mashing . Side hits had to be escaped correctly, as expected.


Q: ( your question here )
If you've got frame questions that can be answered by testing, post 'em up. Try to be as specific as possible about the test scenario (i.e. "use brad K+G on hit to create a neutral situation, then fP+K+G after varying delay against an immediate P+G, see how quickly his ducking avoids a throw", not "what does brad's ducking avoid").

 

Have you tested how the window where throw becomes grab throws? It's been said that the window is 10 frames after a move has hit/been blocked/been evaded, but it would be nice to know for sure.

 

Yeah, thanks for reminding me - that's actually one of the first things I did. Yes, the safety window is 10 frames; I'll put it in.

 

Updated to show that you only need 2 frames per throw escape, including the gap between inputs; as a result, 5xTEG is possible at non-throw-guaranteed disadvantages.

 

About the TE 2 frames thing, do you know how it was in VF4 days ? I find throw escaping harder in VF5, that could be an explanation

 

I don't know for certain how it was in VF4 (although you could probably test by hand - just tap TE really softly until you end up with a 1 frame TE).

Possible that you find throw escaping harder because the input buffer is shorter in VF5.

 

Updated to show why low throw has a more consistent window for beating fuzzy guard than delay throw does.

Sorry about the video quality, I don't have a direct capture setup.

 

KoD do you take requests? I mean, since you have already set this up, could i bother you about some vanessa moves? /forums/images/%%GRAEMLIN_URL%%/blush.gif

/forums/images/%%GRAEMLIN_URL%%/b.gif /forums/images/%%GRAEMLIN_URL%%/b.gif /forums/images/%%GRAEMLIN_URL%%/k.gif + /forums/images/%%GRAEMLIN_URL%%/g.gif cancels

Ever since VF4 when this was they only way i could switch stances in the begining of the round i've always wondered what are the exact total frames for her back charge kick cancels. All 3 of them.

DS /forums/images/%%GRAEMLIN_URL%%/b.gif /forums/images/%%GRAEMLIN_URL%%/df.gif /forums/images/%%GRAEMLIN_URL%%/p.gif + /forums/images/%%GRAEMLIN_URL%%/k.gif

Other than the active inashi frames there's no other info available about that move. <s>Again, total frames would be nice, as well as when the move starts to duck (to see if it will throw clash).</s> I can't seem to be able to get counterhit! I set the AI to counterattack with a mid, there's no yellow flash, counterattack with an OM kick, no blue flash /forums/images/%%GRAEMLIN_URL%%/confused.gif

<s>OS BT /forums/images/%%GRAEMLIN_URL%%/d.gif /forums/images/%%GRAEMLIN_URL%%/k.gif + /forums/images/%%GRAEMLIN_URL%%/g.gif

That move was semicircular in VF4evo, are we sure it's full circular now?
</s>
OS lancers

It's obvious that /forums/images/%%GRAEMLIN_URL%%/k.gif /forums/images/%%GRAEMLIN_URL%%/f.gif /forums/images/%%GRAEMLIN_URL%%/p.gif + /forums/images/%%GRAEMLIN_URL%%/k.gif is still faster than /forums/images/%%GRAEMLIN_URL%%/f.gif /forums/images/%%GRAEMLIN_URL%%/f.gif /forums/images/%%GRAEMLIN_URL%%/p.gif + /forums/images/%%GRAEMLIN_URL%%/k.gif but i'm wondering if the frames are still the same as the ones posted in the VF4evo table (32 vs 40). Is /forums/images/%%GRAEMLIN_URL%%/p.gif /forums/images/%%GRAEMLIN_URL%%/k.gif /forums/images/%%GRAEMLIN_URL%%/f.gif /forums/images/%%GRAEMLIN_URL%%/p.gif + /forums/images/%%GRAEMLIN_URL%%/k.gif still/also at 32 exe? And what are their active/total frames? How about the crimson lancers? They were listed at 13 frames exe, is that still valid? Total frames? And how about the sidestep lancer?

/forums/images/%%GRAEMLIN_URL%%/f.gif /forums/images/%%GRAEMLIN_URL%%/p.gif + /forums/images/%%GRAEMLIN_URL%%/g.gif > /forums/images/%%GRAEMLIN_URL%%/df.gif /forums/images/%%GRAEMLIN_URL%%/k.gif oki

Can Aoi reverse it? I'm under the impression it can't be reversed. I mean, it's the only throw i do, after so many matches and kingo should have reversed it by now. But maybe it's an exact timing thing, so i'd be nice if we could check it.

 

I'll take requests, but not from vanessa players. . . all you need to know is her backwalk speed. . . .






Seriously though, yeah, anyone can ask away, i've got a list of stuff to test. Try to be specific, and ask stuff that's not testable in dojo, though.

44KG - do you mean basically when can she start guarding / how fast of a move is guaranteed to hit her? I can try to test this, but it may be tricky with only 1 set of inputs

43PK isnt that 37 total frames listed in the command list? you can test for yourself if it will duck by doing a -9~-11 move vs cpu set to throw, then inashi. If its anything like wolfs lowpunch cut, it will duck throws.

bt 2k+g should be testable by setting cpu to evade.

the lancers, probably most I could do would figure out what speed of moves they would beat, which wouldnt tell all of the frame data . . . I think catch throws beat attacks if they hit later than halfway through the catch execution?

 

I went ahead and wrote up a quick howto, and released it + my arduino library + a sample sketch that I used for testing lau's kokei frames.

http://code.google.com/p/vf-inputs/source/browse/trunk

Its not really fit for public consumption yet, but there you go anyway.

If you want to submit patches, or sample inputs, just let me know & i can add you to the project /forums/images/%%GRAEMLIN_URL%%/wink.gif

 

KoD, could you do me a favor and test this?
In practice mode, play Eileen w/ the opponent set as Jackey. Set the mode as Counter, do /forums/images/%%GRAEMLIN_URL%%/d_.gif/forums/images/%%GRAEMLIN_URL%%/df.gif/forums/images/%%GRAEMLIN_URL%%/p.gif+/forums/images/%%GRAEMLIN_URL%%/k.gif/forums/images/%%GRAEMLIN_URL%%/d.gif/forums/images/%%GRAEMLIN_URL%%/df.gif/forums/images/%%GRAEMLIN_URL%%/f.gif/forums/images/%%GRAEMLIN_URL%%/k.gif combo, and see what is the last possible frame for the /forums/images/%%GRAEMLIN_URL%%/d.gif/forums/images/%%GRAEMLIN_URL%%/df.gif/forums/images/%%GRAEMLIN_URL%%/f.gif/forums/images/%%GRAEMLIN_URL%%/k.gif to connect as a combo.

I just wanna know how many frames I have for the hit-checking after /forums/images/%%GRAEMLIN_URL%%/d_.gif/forums/images/%%GRAEMLIN_URL%%/df.gif/forums/images/%%GRAEMLIN_URL%%/p.gif+/forums/images/%%GRAEMLIN_URL%%/k.gif connected as a counter hit.

 

I can answer part of that question. it does combo on CH and it combos on NH in open stance for everyone. How many frames you have I donno, I'd imagine as many frames as a CH gives you since you are canceling the recovery into the Kicks. hope that helps a bit.

 

Two ideas of tests:

-Test the speed of Lau's 9P
-Test the time it takes Lei to go in different stance before attacking. Ex: what's the actual speed of Hai Shiki P when you'r in neutral stance. From what I tested it takes more than 3 frames but I'd like to know the exact number. Is it possible to punish -15 moves with hai shiki P,P+K for ex ?

 

Delay timing is much more solid now, will go minutes worth of 1 frame intervals without dropping a frame. Still freaks out occasionally, but I'd feel comfortable suggesting that people build on this if they want to; nothing should change drastically (unless 2p support is borken).

Updated in first post:

Tested lau 9p cause it was straightforward, looks about 22 frames.

Tested some movement cancels that myke and others were asking about, pretty much what you'd expect.

 

Is it possible you could test how many frames it takes for Aoi to get into/out of Tenchi stance? For getting out, I mean just hold G, don't really care about the backturned/evade exit moves.

It seems like a particularly long stance animation, since it's proven completely impossible to do a stance change on wakeup for catching even the most poorly timed okizeme attacks.

 

Thanks for the 9P testing ^^.

 

Another weird question... dunno where else I could ask it. So, my understanding is that two attacks of the same height and damage that both enter active frames on the same frame will cause both characters to take damage, right? My question is, what happens with recovery frames after the attack hits?

So, take for example Aoi's 2K+G, and Jeffry's 1K+G... both low attacks that do 20 damage. Aoi's puts the opponent at +6, and jeffry's puts his opponent at +4. Does this mean that if these two characters hit with these two attacks on the exact same frame, Jeffry will be at +2? Or is it even more complex, and different hitstun lengths on move recovery that would normally would lead to the listed frame advantages will come into effect, and we'll get some weird unique advantage numbers, or does the game actively intervene and the two players would just end up at neutral or something?

 

Hi KoD ! can you test these, please ?

How fast does /forums/images/%%GRAEMLIN_URL%%/f.gif/forums/images/%%GRAEMLIN_URL%%/p.gif+/forums/images/%%GRAEMLIN_URL%%/k.gif+/forums/images/%%GRAEMLIN_URL%%/g.gif Goh execute ?
And how many frames does it take to recover ?

 

With this kind of move that doesn't hit, the tricky part is : WHEN do you consider it "executed" or when does the recover part begins ?
You can count the total amount of frames of the move, from where you inputed it to the moment you can move again, but if you want an exe/recover approach you'll have to define what makes it executed first (ducking state could be a criteria).

 

So go for the ducking state /forums/images/%%GRAEMLIN_URL%%/smile.gif
I've seen there was frame stats for Goh /forums/images/%%GRAEMLIN_URL%%/b.gif/forums/images/%%GRAEMLIN_URL%%/df.gif/forums/images/%%GRAEMLIN_URL%%/p.gif+/forums/images/%%GRAEMLIN_URL%%/k.gif on the wiki commands list, so I was just curious to know the one for his roll.