[ODE] CCD Library (Re: High speed rigid body collisions)

[Patrick Enoch]

I am also using "continuous" physics with ODE. Roughly it is like this:

accept=false;
while (!accept)
{
     savestate();
     step( dt );
     collide();
     if (max_collision_depth > threshold)
     {
         accept=false;
         dt *= 0.5;
         restorestate();
     }
     else
     {
         accept=true;
         dt *= 1.5;
         if (dt > maxdt)
             dt=maxdt;
     }
}

So the objects "stay at the right side"...

-Patrick

On 2. Jun 2005, at 17:22 Uhr, erwin at erwincoumans.com wrote:


> The latest Havok is continuous see their website:
> http://www.havok.com/products/physics.php
> Continuous physics requires time of impact estimation from the  
> collision detector, and a physics mainloop that uses a time of  
> impact queue. A good reference is Brian Mirtichs PhD. Also his  
> paper about Time Warp contains good information about this topic.  
> As long as your physics loop uses conservative advancement to  
> integrate the objects you stay on the safe side. A raycast, a  
> convex cast (see Gino van den Bergens paper using GJK to perform a  
> linear convex cast) or other ways like Stephane Redon's papers all  
> calculate the time of impact. ODE uses penetration depth to recover  
> from missed collisions. However it is possible to modify ODE to  
> allow for time of impact, and you can even have a hybrid system  
> that both handles penetration depth and time of impact.
> I wrote a continuous collision detection library based on Stephane  
> Redon's algebraic method. Also this CCD Library has convex  
> continuous collision detection for both linear and angular motion,  
> using its own GJK implementation. One of the examples uses the ODE  
> quickstep solver. It is not ready for public release, but if  
> someone is interested in this CCD Library you can contact me.
> Erwin Coumans
> ccdlib at erwincoumans.com
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode
>
>
>