[ODE] patch - position-correcting constraints

[Geoff Carlton]

I've recently had tweak of the max correcting velocity and the 
penetration depth layer.  These two settings allow a huge increase in 
the quality of the object collisions, and also seem to reduce the random 
noise in the bounces that makes falling boxes un-stackable.

As a result, the priority of the issue has dropped for me, though I'm 
still curious about any solution that improves the way ode handles 
penetration in the solver.

Geoff

Adam D. Moss wrote:

> Megan Fox wrote:
>
>> The point is that, at present, bodies are de-penetrated by forces. 
>> These forces are in no way separate from others, and stack on
>> eachother with each frame, leading to explosions after stack apon
>> stack apon stack of depenetrating force has its way.
>
>
> That's certainly an issue, I think, but the underlying issue is that
> once that corrective force has served its purpose it still remains
> in effect, which causes both correction-accumulation and the fake-
> bounce effect.
>
>> The patch, in theory, removes that problem, de-penetrating objects at
>> a set, reasonable velocity. 
>
>
> In a round-about way, yes. :)  What it does is remember, within an
> island's step, how much of a body's corrective force is only being
> applied to correct interpenetration, then it removes that much force
> again once the force has affected the position for that step.  So in
> effect it makes the effects of interpenetration-correction not extend
> beyond the current step (any remaining interpenetration will
> cause further correction in the next step, of course), which amounts
> more or less to a position correction.
>
> > This should remove any sort of physical
>
>> explosion due to penetrating bodies, and might even make stacking
>> behavior a bit better (maybe?).
>
>
> It doesn't seem to much affect stacking either way, sadly.
>
>> Apologies on my part, I've been too busy to try your patch - looks
>> great, but I need to keep my codebase consistent for now
>
>
> I understand!
>
> --adam
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode
>
>
>