[ODE] accelerating a sphere with real-world properties / measuring time

[Michael.Rauh at Dornier.eads.net]

Hi,

I'm trying to simulate a rolling sphere with real-world values. Please,
read through my approach and point me to the piece where i am wrong?

I'm applying a torque T to a sphere with radius r and mass m. The sphere
starts rolling along a plane, surface.mu = dInfinity ensures that the
sphere rolls without slipping. I measure the time t needed until the sphere
reaches an angular velocity w.

As we have constant angular acceleration alpha, I can measure alpha = w /
t.
I then compare this measured alpha to the alpha I get from the formula
alpha = Sum of all Torques / Moment of Inertia.

As I understand, the ODE friction model does not apply a damping torque to
the sphere, so my torque T is the only one to consider. Therefore, alpha =
T / 2/5*m*r*r.

But this calculated alpha is exactly 3.5 * alpha(measured), independent of
sphere radius, mass, torque T or desired angular velocity w. I cannot
discover  where this factor of 3.5 is coming from? Is there anything I
forgot?

btw, I'm using dWorldStep with the timestep that has passed since the last
integration (~0.015 on my system).

Please, point me to the right direction. I really need this to get right.

Greets,
Michael