How to handle REALLY small bodies?

[pablonewton]

Hi, I need to know how to handle really small bodies with Newton. This is because in order to manipulate objects with the haptic device I'm using the units of every object must be REALLY small.
I ask this because I did one example where I have a teapot having a convexhull or a NewtonBox as its NewtonBody and a box shaped floor with a collisiontree or a NewtonBox with mass = 0. When I start the application the teapot is placed a little above the floor and it falls on it, but starts to bounce and move, and in the end it falls through the floor. I think this is because of the dimensions of the bodies that are too small, having dimensions of their bounding boxes like these for example:

Teapot : widht_x: 0.014333, height_y: 0.011242, depth_z: 0.022942
Floor: widht_x: 0.155701, height_y: 0.017946, depth_z: 0.088241

1) So, Is there a way to tell Newton that the units used are really small in order to avoid this tunneling problem?

2) I also I noted that the teapot falls really quick, but if I scale these dimensions to make them bigger, the teapot falls a lot more slowly.
Do I have to scale the values of the forces and masses for every object so they don't move so fast? (I set the gravity = mass*9.

3) I also set this:
NewtonSetMinimumFrameRate( newton.GetNewtonWorld(), 300 );
in order to have more accurate collision detection, is that correct?

I gave both bodies a wooID material like the tutorials.

Thanks in advance.

[Stucuk]

If your entire world is rearly small, then you can have the physics scalled up as if it was a larger world and then scale down the positions/whatever when needed(i.e rendering).

[JernejL]

You can try using the double precision library if that helps.. but i really suggest for single precision that you use object sizes that best use range from 0.01 to 4096 (also -4096), usual set-ups are 1 unit = 1 meter, the floating point imprecision is simply not workaround-able in any other way, also try to avoid having ridiclously huge and ridiclously small objects interacting..

[pablonewton]

You can try using the double precision library if that helps.. but i really suggest for single precision that you use object sizes that best use range from 0.01 to 4096 (also -4096), usual set-ups are 1 unit = 1 meter, the floating point imprecision is simply not workaround-able in any other way, also try to avoid having ridiclously huge and ridiclously small objects interacting..

Thanks Delfi, but with regards to the double precision library:

a) Does it fix the tunneling problem I mentioned?
b) Can I use smaller values with this library (in Newton) than 0.01?
c) How does this library can be used wih Newton?
d)...Where can I get it?

Thanks again.

[JernejL]

You can try using the double precision library if that helps.. but i really suggest for single precision that you use object sizes that best use range from 0.01 to 4096 (also -4096), usual set-ups are 1 unit = 1 meter, the floating point imprecision is simply not workaround-able in any other way, also try to avoid having ridiclously huge and ridiclously small objects interacting..

Thanks Delfi, but with regards to the double precision library:

a) Does it fix the tunneling problem I mentioned?
b) Can I use smaller values with this library (in Newton) than 0.01?
c) How does this library can be used wih Newton?
d)...Where can I get it?

Thanks again.

A: use continous collision
B: i haven't tested, but seriously you should scale up your sistem
c: what?
d: get it here, double precision libraries are included. viewtopic.php?f=9&t=4922