Object With Up Vectors Flipping 180 & 360

[dylanmaxwell]

I've observed that objects with up vectors are flipping over as they animate. It either flips 360 or 180 on the other two axes. This is fine in maya (even with motion blur) but I'm rendering elsewhere and it's causing a huge motion blur on those frames.
my questions are:
why?
and how do i work around?

[gmask]

well I guess the I'd start by figuring out why they are flipping otherwise you could try to counter the flips as they happen

[joojaa]

Yes thet because the up vector assigns a solution for the virtual plane your rotating about whenever you move over the up vector it flis around the plane (or actually it spins VERY fast), and whenever your vector moves in the quadrant it will issue a flip too.

However this does not seem to be a issue with the aims flip but rather, when you turn stuff into matheamtical operations that depend on normalized vectors, or rotations in transfrom stacks it matters wether yyou go form 0 to 179 or 360. Thing is that a aim works on each frames data so it DOES NOT KNOW what quadrant is in use. You should bake your animation down to keys and run a euler filter on the curves. before export. The problem with this is that it does not know if you rotated more than 180 degrees in one frame.

Maya nodes in general dont account for their last position for anything all calculation works as if only 1 frame existed.

[dylanmaxwell]

Ok, so I understand the why but the how is still an issue. I've been working on a perl script to fix the output but maya seems to be able to do it on it's own as maya software and mental ray have no trouble with huge motion blurs.
I can bake keys and then manually go through and fix sections but I really need to automate this output

[joojaa]

QUOTE(DylanMaxwel @ 01/03/09, 06:27 PM) [snapback]298087[/snapback]
but maya seems to be able to do it on it's own as maya software and mental ray have no trouble with huge motion blurs.

They interpolate the solutions manually, so what happens is the rendered always sees the difference between the current and the previous matrix. They never gather up this problem because they never deal in rotation in this manner. But rater they store the result and the previous result. As a downside they may need to supers ample the scene if you need some rotations that are more than one time around.*

However there IS a problem, the system has hard time seeing when one frame rotates many times around. Technically you could chek if the datas coming in form animation keys if so use raw values (or better do a predictiin based on how the connetion node normally works). Else just do the interpolation manually. Because nodes don't necessarily know of the difference. Bacause many of theese solutions are matrix based. That are just converted back to eulers. Mosty they just super sample the scene frames.

Any noneuler calulations DO NOT have visibility outside a strict one slice of the rotational space. Quats, axis angles and matrices only have one rptation to work with.

QUOTE
I can bake keys and then manually go through and fix sections but I really need to automate this output

you can automate that too.

• BTW, its not just mentalray that works like this MtoR, gelato and liquid do too. Basically thsi si the only safe way. But even it fails whan you need those wild rotations. This si also way simper for your export.

PS you should be using the api.

[dylanmaxwell]

ok, thanks. I've observed that the flips also sometimes flips the direction of animation on non flipped axes.. Of course. Also, the invert of the curve pivots on the curve value at the moment of flip... ugh.

[gmask]

QUOTE(DylanMaxwel @ 01/03/09, 10:20 AM) [snapback]298093[/snapback]
ok, thanks. I've observed that the flips also sometimes flips the direction of animation on non flipped axes.. Of course. Also, the invert of the curve pivots on the curve value at the moment of flip... ugh.

are you using a motionpath?

[dylanmaxwell]

QUOTE(gmask @ 01/03/09, 01:30 PM) [snapback]298094[/snapback]
are you using a motionpath?

At times yes, but this occurs any time i use an up vector.

[joojaa]

Yes they are easy to predic for path animt it happens when the curvature moves trough the direction of the pole. This is relatively easy to automatically fix (because its EASY to know where the flips will ooccour, its just a curve intersection with a offset version of itself). Tough it don't help if you use path flow.

Technically the node could be rewritten (its not hard, its not a terribly complex node) for it to work as users expect. But for philosophivcal point of view its better that the nodes deal with vanilla mathematics than hide specific magic solutions beneath them. Its easier to debug, besides thats kindof the point fo why you choose the up vector.

The mathematical idea is just simple cross product between up and along for the remaining 3 vector and then 3rd and along for the new closest possible up. Then they are normalized for a rotation matrix. This matrix is then turned to eulers. But ONLY eulers have the capacity to contain more than one way around rotation values.*

If it wasn't simple then debugging other things would become impossibly hard. And we would have a 3ds max sort of experience. It wouldn't make sense to call things things they are not, just throws people off. All of theese can be countered but the idea is t keep the poles constrained in way that you never move to those bounds.

But like i said the only way to avoid filiiping i to do the intrpolation yourself. Using either quats, mtarices or or axis angles. thsts the only way. After all 0 IS the same rotation as 360, the nodes deal in calculating output not interpolating them.

*to offset this not all rotations are possible to interpolate in all circumstances easily in euler space. Reality does not have such things as rotation 2 times around, it just has a psotion and angular speed velocity etc. But its allways jsut that position.

[dylanmaxwell]

I see and agree. Still, an "I don't care, do it wrong" button would be nice. maybe on the bake keys button.
thanks for all your help.
-dylan

[joojaa]

QUOTE(DylanMaxwel @ 01/03/09, 09:09 PM) [snapback]298099[/snapback]
I see and agree.

Thats what the euler filter option in the curve editor is for, its a please make these curves make sense to me kind of operation.

Thing is theres nothing simple about 3d rotations. They can be done well in so many ways but they all lose something about whets well in the other ways. Thing is intuitive understanding on what happens was just chucked out the door the instace you start to deal with 3d rotations. Quats come closest to being mathematically perfect descriptors but teir only drawback is tht the values don't make sense to humans. Wich form my perspective is a terribly bad downside.

Thats why we all read human input in either eulers or axis angles and convert them to matrices and quats for calculation. It snot that eulers are understood inherently better, usually people just think they understand them better, bu they make sense when you only rotate one axis at a time.

QUOTE
"I don't care, do it wrong"

you arent so happy when somebody asks you to fix problems cause by this. They are equally common as this problem. Theres no such thing as pefect, just imperfect you can deal with. this is a sort of mac engineering thing, lets do what most people need 90% of the time and lets not make it possible for thm to do the 10%.

Maya is about the 10%, if you arent interested in it don't use maya.

[dylanmaxwell]

QUOTE(Joojaa @ 01/03/09, 02:21 PM) [snapback]298100[/snapback]
Thats what the euler filter option in the curve editor is for, its a please make these curves make sense to me kind of operation.

Oh i see, I've never noticed that before. Will this cause an inaccuracy in certain situations? there are no fast rotations in my scene. It looks like what i need for outputting my baked cam.

[joojaa]

QUOTE(DylanMaxwel @ 01/03/09, 09:44 PM) [snapback]298101[/snapback]
Oh i see, I've never noticed that before. Will this cause an inaccuracy in certain situations?

Well to be honest, matrix to euler conversions produce wierd results sometimes. And assuming the renderer doesn't work on those eulers it would work right. So yes the curves will work out but wether or not they do the spirit of the actual move is another thing entirely (its not that other methods do more better but atleast they can be more predictable).

However if it does then the only way to get halfway decent results is to turn the 2 euler values to quats and interpolate that. Because while eulers can describe all rotation ina 3d space they can not properly do all possible interpolations. So no matter how you put it using eulers as a basis for rotations for interpolation is a bad thing. Theres nothing you can FIX about this its inherit property of the mathematics. But mostly it works out okish as long as you dont plan to blend the values with other stuff, or want to do some real manufacturing.

[joojaa]

So to recap you can easily remove the flip, but the fact that how eulers dont work out right is harder to fix. But this problem happens WAY before the flip occurs.

[dylanmaxwell]

OK, thanks again.
Sorry to make you whip out your mac user metaphors.

[joojaa]

I have nothing against mac. Its a good paradigm, it makes a awfully lot of people happy, it makes them feel worth the same. Life is about perceptions if somebody feels good at that then its the way to go. Offcourse the osX has inbuilt seeds of the opposite inbuilt. But you see this same employed well on iphone, ipods etc. Unfortunately its not aimed for me, or my world view*.

*its a tradeof