Newton 2.0x Archemedia Open Beta

[collerblade]

Did anybody able to download newton form this link http://www.newtondynamics.com/downloads ... n-2.19.rar? the file appears to have bad header and im unable to properly extract it.

thanks.

Yes. I did & it worked for me. Try again. Or try a different internet connection.

[shybovycha]

publicENEMY, maybe you'd like to wait for a newer version?

[rvangaal]

I've been testing with Newton 2.10 and 2.19 concerning a problem (bug?) in the collision tree.
I have a quite long track (400 meter) with a wall that runs alongside it. When creating the collision tree with optimize enabled, I can drive through the wall. When disabling optimize, it works correctly.

An image of the optimized physics mesh is viewable at http://racer.nl/temp/bug_collision.jpg

It seems that optimization takes co-planar triangles and combines them, but it does this so aggressively that very long polygons are generated, and the collision (of ConvexHull, the car) with the tree fails at most points. Disabling optimize results in small collision tree polygons and then the hull collides correctly.

We've tried tesselating the original track (wall) mesh, but that is just optimized out (Tried it in Newton 2.19). Any ideas what else I can try?

[Julio Jerez]

That could be right, I maybe have to limit the size or apect ratio of faces in teh optimization face,
My data mesh sample bases do not represent real world game case, and this is one of those cases.

I am going to revisit the Collision trees this weekend to make the build process more efficient, because many people had complained that it too very compared to what is was in 2.09
The lastest optimization made it too good for its own good.

anyway can you use non-optimizid faces until i publish 2.20 this weekend?


I will add
- make it do no create sure faces with too long aspec ratio
- make teh build process faster.

It is possible to get one of you Track mesh for me to test this?

[rvangaal]

I don't have a tight deadline, so 2.20 would be fine, great that you're putting the time in.
Most of time I don't have any problems, just this very long mesh.
I can provide an ASE file for example; the original is in 3D Studio MAX (2010?). Which file format would you like?

[Julio Jerez]

ASE or Max are fine.

[Julio Jerez]

Tell me something,
are the long indvidual faces very long to begin with or do they become long as result of the optimization in the collision tree?

[rvangaal]

The tesselated version had quite normal triangles (no big aspect ratio); the optimization made them long.
I've sent you the .max file by private message.

[Julio Jerez]

What I mean is if the non tesselated version has long thin faces to begin with.
because if the faces are long to begin with, The only solution is to clip then until the apect ratio is satifactory.

It is not that the face is big the is the problem is the long aspect ratio.
Bu I do not think I shold cut faces becuse some people use collision trees for other porpuses.

[Julio Jerez]

Ok I finally have the mesh converted to a Collada file and I and loading it in the SDK
Tell me something are the collision fail on thr frnce polygon on the side of the track? are the misses consitent?
I loaded it in the SDK and while I can see the fence being way too long, I did not see the bug when testing it with few collision shapes.

So I am wadering if this is a intermiten bug or it is a failure because a break down of the Math in the collison routine.
what I am tring to ask is if the collision is consistent when optimization is off, because even with optimization off I see very, very long slivers faces too.

when I optimize the mesh I am getting gace that are 890 unit in length and less than 10 in width,
This is ceratlly very bad becaus eit leat to poor contacts and slow convregence,
I definilly need to fix that. on the brigh side the solution will be a wins, in all aspects.
The only draw back is that I will only add it to optimiza meshes, not optimized mesh will remaind the same as the application pass it.

[rvangaal]

The collisions fail with a convex hull with the optimized tree only. Also, at some points the collisions do work, it has problems near the start of new polygons. Almost as if one of the two polygons that makes up a side wall is not present (or detected). It also depends a bit on which side you are.
Near the start (near the big gray 'parking area') I can drive a car through the wall to the outside. Then trying to get back on the track it does detect collisions, until I drive towards the next collision polygons, where I can suddenly get back on the track. The problem doesn't exactly reverse there, it is more that at some point a bit further I can collide again.

The original mesh has pretty long polygons too, but without optimization I have no problem. Not sure how hardcoded the aspect ratio at which you reject optimization should be, perhaps best configurable.

Meshes that already have very long polygons are bad anyway, also for graphics painting it's not ideal (culling for example), so I wouldn't worry about that too much.

[JernejL]

rvangaal: trimesh is one-sided only, polygon direction matters, so make sure you are sending proper data to newton to build the trimesh.

[rvangaal]

I know; without optimization it does work, and has worked for many tracks sofar. So it's definitely a one-off in a row of many succesful trees.

[Julio Jerez]

I know; without optimization it does work, and has worked for many tracks sofar. So it's definitely a one-off in a row of many succesful trees.

That's one off too many in my book, and we cannot have that can we?

I run the test and this is what I found
-no optimize the worst aspect ratio is 2294 (face is 458 long by 0.2 wide)
-optimized the worst aspect ratio is 2544 (face is 510 long by 0.2 wide)

Those are really, really bad figures by any standards.

The engine should coupe with easily with aspect face with aspect ratio equal or less than 100,
But I never run a test on that I will limit it to 50 arbitrarily.
There are 52 faces with height than 50 aspect ratio.

I think I will get this going tonight.

I will only do it for the optimized mesh option, no optimized option I leav to teh the end app to pass well behave meshes.

I am always reluctant to expose that kind of non physical parameter, because it create lot of confusion for other people,
since the mesh is working with much higher that 50 aspect ratios let us put that as a limit an see what happens.
On another hand this is also a very special mesh with very, special characteristic it may be a good ide to make optional.

what I do not like about optional parameters is that you never know what to use it. so let us try uncoditional 50 apsect ratio limit.

[Julio Jerez]

I know; without optimization it does work, and has worked for many tracks sofar. So it's definitely a one-off in a row of many succesful trees.

Ok I added the aspect ratio parameter to collision tree build.
Please try and let me know if it works for you.
get SDK 2.20 from the download.