Coordinate system flips when pitch crosses 90deg?

[bernie]

we realized an odd behavior in observer coordinates: When the observer moves on a continuous trajectory that curves upwards, as soon as the pitch crosses 90deg (i.e., as soon as the observer is upside-down) the observer coordinates seems to flip upside-down as well (resulting in a change in heading, pitch, and roll coordinates)...
Is that a know feature (is it documented anywhere)?
more importantly, is there a way to switch it off?

We're trying to compute hoping vectors and the whole things get's messed up currently when the coordinates have a discontinuous jump .

Thanks for any advice!

Bernie