[ODE] Large mass and accuracy

[Billy Zelsnack]

>> So I guess all values should be centered around the same
>>magnitude in order to keep the losses inaccuracies low

>That's what I've experienced too, the ODE doc says the same.

>> But then I'd have to adjust

>I have the same problem. I need
>to get a realistic result with realistic dimensions of input values.
>But what do you do when you want to connect a wheel with 50 kg
>mass to a car/tank body with 50,000 kg mass and keep a stable
>simulation? Didn't find a proper solution so far.

The problem is not floating point accuracy. It is a slow converging 
solver when high mass ratios are present. The key word here is ratios. 
Imagine a perfect stack of two spheres. The top sphere is really heavy,

the bottom one is really light. Whenever the light sphere interacts
with the ground, its inertia is instantly stopped. Whenever the light
sphere interacts with the heavy sphere, its inertia is instantly
stopped. However, the heavy sphere keeps on trucking. The light sphere
takes only a tiny dent out of its inertia, but not enough to stop it.
This is why constraints sink or bend. Increasing the number of
iterations will help out, but that only works to a point and many times
is not feasible. Fixing the sinking problem can be done via trickery.
One method is explained in the paper 'Nonconvex Rigid Bodies with
Stacking'. Fixing the problem with other constraints isn't so easy.

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com