[ODE] Mapping to OpenGL axis system

[Ruud van Gaal]

Hi Russel and the group,

I'm trying to create an ODE version of my racesim, and due to my lack of 
insight on matrices, I run into some trouble with coordinate systems. My sim 
uses the OpenGL axis system, meaning a right handed Y=up, X=left, Z=out the 
screen. For the cars, this ends up as the Z axis point from the seat of the 
driver towards the front bumper, X left out the door, Y up.
Now I don't know exactly what ODE uses, but it seems like left-handed, with Z 
going up?

After modifying my rigid body class to create an ODE dBody, for painting in 
OpenGL I had to map ODE's axis system to my (OpenGL) axis system. As it turned 
out, inverting the quaternion from ODE worked using myQuat.w=odeQuat.w, 
myQuat.x=-odeQuat.x, myQuat.y=-odeQuat.y and myQuat.z=-odeQuat.z. So I had to 
negate the xyz members. Then I derived my own 3x3 matrix as before ODE, and 
with this matrix the car started painting in the expected orientation again.

However, when applying forces and torques using dBodyAdd[Rel]Force(), I seem to 
have to do an extra mapping. Adding body forces seem ok with dBodyAddRelForce() 
and I can just pass force.x,force.y,force.z (although I'd rather pass a pointer 
to these 3 floats, but that's another story).
Adding torque values are weird though. When using dBodyAddRelTorque() with my 
(OpenGL-based) torque.x/y/z, things go bad quickly. Using world coordinates and 
passing them as body coordinates, I seem to get closer with 
dBodyAddTorque(bodyID,-torque.x,torque.y,-torque.z), but then steering left 
pushes the car around right and vice versa. Adding some negations here & there 
make it graphically correct, but the car is very sensitive to spinning and I 
feel my coordinate system hack just doesn't work.

Does anybody have a good idea about how get the OpenGL and ODE's axis system to 
work together like it should? It seems I must convert from right- to 
lefthanded, but as quaternion rotations are done presumably assuming a 
lefthanded system, my AddForce()/AddTorque() calls go wrong.

Thanks for any ideas, I've spent multiple hours in this already and getting a 
bit nuts and I see a maze of negations coming up, which is definitely NOT good 
(and pretty endless too).
I want to avoid switching all MY stuff to lefthanded, since that would make the 
conversion bigger than it's supposed to be.

Ruud van Gaal