Air65 — The Nose-Up Takeoff, Traced to the Level Reference

Finding

On the last few launches the craft tilted back as it lifted off, and a physical inspection found nothing. The logs show the pitch level reference stepped by about 6.5° between pack 6 and pack 7 and stayed there. Angle mode held a standing pitch correction while the craft sat on the ground, and it released into the airframe when the craft broke ground.

The fault

With the craft armed, stationary and sticks centred, the accelerometer should read level:

Logs	Ground pitch	Ground roll
`btfl_001` … `btfl_024` (packs 1–6)	−0.9° … +0.9°	~0°
`btfl_025`	+6.11°	−0.06°
`btfl_026`	+6.19°	+0.00°
`btfl_027`	+6.50°	+0.00°
`btfl_028`	+6.50°	−0.14°
`btfl_029`	+7.55°	−0.92°
`btfl_030`	+6.31°	−0.11°

Pitch only; roll is untouched. The step lands on a power-cycle boundary and persists across a further one. The gyro is not implicated: it reads exactly 0.00 on all axes whenever the craft is stationary.

Why that tilts the craft

Angle mode does not wait for takeoff. Believing the craft to be tilted, it corrects while still on the ground, and with pid_at_min_throttle = 1 the correction reaches the motors at idle:

Log	rear-R	front-R	rear-L	front-L	Split
`btfl_025`	3 017	10 011	3 274	10 022	+104 %
`btfl_026`	2 996	9 960	3 153	9 858	+105 %
`btfl_029`	2 982	9 592	4 011	10 547	+97 %

Before log 025 that split is within a few percent. Raising throttle carries the asymmetry up the ramp: 50 ms before liftoff in btfl_030 the motors read 7 150 / 19 233 / 7 467 / 19 800 RPM. Thrust goes with RPM squared, so that is roughly seven times the thrust on one end of the airframe at the moment it becomes free to rotate.

What is not wrong

Gyro: zero bias on all axes when stationary, in all 31 logs.
Roll axis: ground roll within 1°; left/right balance normal.
Hover balance: front/rear split settles to about 3 %, against 2.3 % in the earlier review.
Motors: RPM per unit of command within ~2 % across all four in every clean flight.

Fix

The accelerometer was recalibrated on a level surface over USB and saved.

	`acc_calibration`	Attitude on a level surface
Before	`-2,-7,36,1`	pitch +5.0°, roll +0.0°
After	`-179,-8,-25,1`	pitch −0.8°, roll −0.4°

Verified stationary for 90 s (0.84° of drift, no stepping) and again after a reboot. acc_trim_pitch and both align_board_* values were 0 throughout; only the calibration offsets changed.

Root cause is not settled. The data does not separate a wrong stored calibration from a craft that really was resting 6.5° nose-up on its launch surface. The evidence leans to the calibration — a pitch-only offset, constant across six arms and two power-ups — but it is not proof. The check: the first second of the next session’s log. Ground pitch near zero and a small idle front/rear split means fixed; 6.5° returning on a known-level surface means the board.

Betaflight reports attitude pitch with the opposite sign to asin(accX / 1G). Verified live against MSP_ATTITUDE: accX −175 counts reads as pitch +5.0°.

Housekeeping

The flash was full (16 777 216 bytes) when pulled, so logging had already stopped. Erased.
motor_kv = 1960 is still wrong against ~27 000 RPM observed on one cell. Harmless.
dyn_notch_max_hz = 400 still sits below the ~455 Hz motor fundamental.