[ODE] Pushing through collisions

[Nguyen Binh]

Hi Andrew,

AA> Cases like that are impossible to correctly fix in the general case without
AA> using further information.
    Yes. You are right! In this case, just geometric information is
    not enough.
AA>  A penetration system that used the objects
AA> velocities could for instance potentially "weight" the axis seperation tests
AA> to have it tend to give axis that are as close to the negative of the
AA> direction of movement as possible.  Havent really looked into it much -
AA> finshing off some cylinder code myself -)  Without context and with high
AA> speeds, or "high" rotations you'll always be able to find incorrect
AA> seperation determinations ( ie your radial response was correctly the
AA> smaller seperation ).  I would be wary of hacking in fixes to try to correct
AA> for this - as crystal-ball methodolgies tend to break down at the worst
AA> times -)

    I used the different velocity vectors of two objects to make a
    guess on depth normal.
    
AA>  - Andrew Aye

    Nguyen Binh.