SV: [ODE] Particle/cloth/skeleton (was: Re: Choleski factorization)

[Joakim Eriksson]

> >In practice, a
> >dozen iterations seems to be more than enough, at least for simple
> >systems.
> the skeleton used in hitman was very simple(no neck, waist 
> joints etc, no twisting etc...)
> and it required about 10 iterations if im not wrong. Further the
> interactions i have seen in
> the game were at very low speed(just moving dead bodies around at low
> speed), which makes me
> thinking that for high speed interactions many more iterations are
> needed.Probably

This is not true. The constraint system was activated as quickly as
any force (Bomb/bullet) hit the character so there was some large
forces involved sometimes. I was at GDC and heard his talk (He's
a very funny and good speaker by the way) and there he showed a
level with a bunch of characters and then he set off a big bomb.
In the middle and they all flew away. Looked fun.

Something that he also showed there was a 2d watersimulation that
used the full navier stokes equations (So no it wasnt a simple surface
simulation). It all ran in realtime and it looked just so amazingly
good.
He didnt say much about it but he did mention that it used the
relaxation method also. To this day I still can understand how he
managed to do that. I wrote to him a few days after and got to known
that he had gotten it to work in 3d also. That was many years a go now.
I had hoped they would show some of his new stuff in Hitman 2 but 
that they didnt...

> the best way of evaluating such system is to implement it. 
> Currently im using an direct LCP solver +
> iterative LCP solution + featherstone. months ago i implemented also a
> cloth simulator that among many things
> also enforces the length constraints the way you just said. From the
> performances indications i have the particle method does not 
> seem to be a good substitute for more "classical" methods.
> But it could be just my implementation.

It can be a good substitue. Specially for the normal constraint part.
Because its very fast at solving constraints (and I dont mean just 
simple distance constraints). Its also nearly impossible to blow up.

> further im not quite sure at how friction would work on a rigid bodies
> modelled as pseudorandomnly
> distribuited particles. We are talking about generic systems not only
> simplified skeletons colliding against a static environment.
> we may want skeletons to collide against each other or against other
> objects, etc..After a collision as been detected the 
> collision force can
> only be applied at the available particles. not quite sure about
> the all thing and clearly this part is missing in the Jakobseen paper.

When it comes to the skeletons and collisions detection he does metion
that he uses capped cylinders for it. Then he probably just displaces
the correct particles using the collision point.

> Hitman doesnt' have generic rigid body simulation if im not wrong.

In the paper he states that he models rigidbodies using a tetrahedron.
Its hardly not used in the games and I realy dont know why. But in one
place I remember they had a trashcan that you could tip over and 
shoot at. It was surprisingly fun =)

> however more detailed informations on the particles based 
> approach can be found on the Jeroen M. Wagenaar Phd(he worked on
> Hitman 1 and 2 if im not wrong).

I have been searching for this paper but I cant seem to find it
anywhere. Do you known where I could get a hold of it?

/Joakim E. - http://www.planestate.net