Page 2 of 2

Re: testbuild 2005july19: fix for random collision problem ?

Posted: Tue Jul 19, 2005 10:37 am
by fobsta
The physics system seems to be shutting off rather abruptly.

In the multiple object demo scenes you bundle with this new build, after the objects have gone through all this spectacular collision and are settling down to their rest position, they all stop moving at exactly the same time.

It might look more natural if you stagger their deactivation?

Posted: Tue Jul 19, 2005 6:52 pm
by pildanovak
erwin: I'm not sure about the sex of the monkey, I thought her name is Suzanne(?). :wink: but I'm czech and am not really an expert in the monkey things.
I'm going to test the build...

Posted: Tue Jul 19, 2005 7:31 pm
by pildanovak
the monkeys don't jump anymore.

I realised another thing- if stacking a high pile of objects and then letting another fall on them from height(so that they can get in rest state inbetween) results that after collision some of the objects flying around stop in the space.

another issue- the damping value is implemented inverted. 0 means high damping, 1 no damping.

Posted: Thu Jul 21, 2005 3:51 pm
by erwin ...

This one has no deactivation, which means objects stay active forever. This should stop the objects from deactivating too early. But in the future, tuning the deactivation settings will be probably better.

Please report feedback on this one too. Build with Visual Studio 8, 2005 Beta 2.

Posted: Sat Jul 23, 2005 10:44 pm
by Saluk
Yes, this version where the physics never stop seems to work really well, and it DOESN'T have the jittery problem most physics engines have if you don't kill the evaluation. I don't know how you did it!

The box-on-sphere test slowed to a crawl
In the PolyTope test, it seems that sometimes they rest somewhat inside each other.

Posted: Sun Jul 24, 2005 12:29 pm
by erwin
first of all, thanks for the encouraging words.

>The box-on-sphere test slowed to a crawl
if you make the bounds 'sphere' it should fly.

the sphere is tesselated, and I didn't add bounding volume tree support inside the mesh, so it brute force does way too much work. also the penetration depth calculation is very rudimentary (but it works, slowly).