Crass bug fixed leads to the complimentary filter smoothly tracking the noisy angles from accelerometer:
The complementary filter almost certainly needs refinement – I think the level of smoothing is too high. At the same time, I suspect I can drop the dlpf from its current 20Hz. Together that should help.
But at the same time, a test flight still showed Phoebe drifting forwards in the real world due to sensors reading backwards movement the sensors’ world.
P.S. Interestingly, the angles suggest Phoebe thinks she’s level, which would then suggest the sensors are misaligned compared to how they were when calibrated. Must check to see if they’ve worked themselves loose since calibration – that would certainly account for pitch drift, especially as the pin layout would work itself loose along the X / pitch axis,
P.P.S The bug fix is now on GitHub.