Hehehe - my bad!
As I have been discussing with a couple of people who have been testing the code, if you do not have the correct accel attached, you will need to fake the outputs of the accel subroutine with x=0 y=0 and z=1024.
The code will then have the illusion that it's the right way up, and it's not falling/climbing.
I hope this should fix it - I have to confess I have tested the code without RC, and without Serial input, both work fine, but the removal of the accel causes big maths issues in the mixing of the control data.
I hope this helps, there will be some slightly tidied up code going live today. Many little changes, some useful, some just to remove things that have caused confusion.
All the best,
Ed