[ODE] contact joints / attaching them

Nate W coding at natew.com
Sat Oct 11 12:27:50 MST 2003


On Sat, 11 Oct 2003, gl wrote:

> > It isn't.  Collision detection is concerned only with the interpenetration
> > or contact of GEOMETRY objects.  When a collision is detected, a contact
> > joint must be created between the colliding objects.  After that, it's the
> > DYNAMICS code that takes the timestep into account when determining how
> > much force to apply to the BODY objects.
> 
> Yeah, but the amount of penetration determines the force.  And the amount
> will be different with different timesteps - ie. the timestep is relevant to
> the collision results indirectly.

Yes, what you say is completely true... but it does not have any bearing
on the Flavien's (correct) assertion that collision detection functions
are completely independent of the timestep.

The CD code only computes the penetration depth, and it doesn't need
timestep information to do that - it only needs to know where the geometry
is currently located.  As far as the CD code is concerned, all objects are
motionless.  The timestep does come into play in the dynamics code, but
that's *after* the CD code has finished executing.

I just realized that I neglected to answer the other question Steve asked
in the message I was replying to:

> How can the system tell whether a ball thrown at a wall - which is on
> one side of the wall at the start of the timestep - and on the other
> side at the end of the timestep?

The system cannot tell when this happens.  This is sometimes referred to
as "tunneling" and it's a known issue.  Small fast-moving objects can
effectively pass through other small (or thin) objects.

You can get extra-high scores in the dismount games if you can find a way
to, for example, get the guy's arm to pass through the roof of the truck.  
This results in contract joints that try to push the arm and body apart...
Then the shoulder joint tries to bring them back together again, but that
just bangs them into the truck, repeatedly.

-- 

Nate Waddoups
Redmond WA USA
http://www.natew.com/




More information about the ODE mailing list