[ODE] Pushing through collisions

[Jon Watte]

     Force is supposed to be exerted by the ODE contact constraints is
    to be determined by the solver base on penetrate depth. So the
    penetration depth is trivial, i.e collider's jobs.

Let me be more specific in my questioning: Exactly how big does the 
solver make the forces, based on penetration depth and mass? Does it 
also look at other values? (linear velocity, for example) Where in 
the code can I look at this in detail? I've found the general area, 
but haven't had the time to study the constraing solvers in detail, 
and am looking for an accelerated start.

Also, my problem is not a temporary interpenetration, it's that by 
pushing up against the block, I become more and more embedded into 
the block; there's penetration all the time (I use ERP < 1).

      Those problems are raised by not robust colliders. I think of the
     the most "not robust" part of ODE is colliders.

It generates the collisions and depths just fine. Box/capsule is not 
among the most complex colliders, and the code appears to do the right 
thing (after I corrected for the denormal case :-)

     (originated from Croteam's SE 2.0 engine). There is case in
     cylinder vs sphere I have to exaggerate the depth value to make
     the huge object didn't go through small one. Let's wait and see

That tells me that something else is wrong. You should not need to 
exaggerate anything -- thus, I think it's the joint constraints that 
are not correct, rather than the collider. Hence, my question about 
how it's supposed to work.

> JW> Does ERP or CFM affect this, and if so, how?

>     Yes!

>     <From ode doc>

I already read the documentation, and I use an ERP of 0.9 (IIRC). 
The question was whether ERP affects the constraint forces in any 
way other than saying, "I'm D deep into this collision, and the 
body is at mass M, and time step T, so I'll apply force that'll 
accelerate the body ERP*D units in one time step". It sounds like 
"no" is the answer.

Note that once you have penetration, if you correct that using an 
impulse, you will have velocity going OUT of the object, so there 
will always be some amount of bounce. I am under the impression 
that ODE applies its corrections ONLY using impulses -- is that 
correct?

Perhaps the answer is as simple as the slow squishing of the avatar 
into the box is because of a poor CFM/ERP constant choice; I'll have 
to go back and re-look at it (especially CFM). However, I'm still 
interested in hearing about experiences in moving objects that 
collide, rather than just applying force. I would imagine that if 
you penetrate very far in a single time step, the force would be 
such as to generate a pretty hefty speed going the other direction, 
and thus actually generate "bounce" even if I didn't want it. Having 
a rule that, for a contact, the object moved by (1-ERP)*depth and 
had an impulse applied by ERP*depth is that I'm going to try next.

Cheers,

			/ h+