[Bug] Copy Trims (fixed-wing) does not preserve elevator position, surface ends at a different angle after zeroing TX trims

[orrerie]

Describe the bug

Environment

• PX4: 1.17.0 alpha (git 3e396f65e5e1c343ec6ad09fd761e27a80b7275)
• Airframe: Fixed-wing, control allocation (CA_SV_CS*)
• GCS: QGroundControl
• Connected surfaces: elevator (pitch), rudder (yaw)
• Flight controller: Cube Orange Plus

Summary

Copy Trims is expected to be transparent: after pressing it and zeroing the TX trims, the control surface should remain at the same physical angle it was trimmed to. Instead the elevator ends up at a different angle.

Steps to reproduce

1. Apply elevator trim on the TX. Elevator deflects to 23° up, RC channel reads −23%.
2. Press Copy Trims. Elevator jumps to 44°.
3. Set TX trims to 0. Elevator settles at 41°.

Expected vs actual

• Expected after step 3: ~23° (the angle it was trimmed to).
• Actual: 41°.

Notes

• After the procedure, TRIM_PITCH = 0.23, matching the −23% channel offset that Copy Trims captured.
• Step 2 (23° → 44°) looks like the offset is applied twice: the live stick signal still carries the trim while TRIM_PITCH is simultaneously written with the same offset.
• Step 3 should return the surface to the originally trimmed angle once the live stick offset is removed, but it stays at 41° instead of ~23°.

Parameter diff (before vs after the procedure)

Param	Before	After
TRIM_PITCH	0.0	0.23
RC2_TRIM	1500	1500 (unchanged)
CA_SV_CS2_TRIM	−0.05	−0.05 (unchanged)

Copy Trims writes the trim offset into TRIM_PITCH rather than the RC or control-surface trim.

Question

Is this expected behavior? If so, what is the correct way to use Copy Trims so the surface keeps the trimmed angle? If not, Copy Trims should preserve the physical surface position after TX trims are zeroed.

Flight Log / Additional Information

No response

[SofianElmotiem]

Hi, looked at the do_trim_calibration code and the step 2 jump makes sense: the moment Copy Trims writes TRIM_PITCH = 0.23, the rate controller is summing both the live TX trim offset and the new TRIM_PITCH at the same time, so you briefly get double the deflection. That resolves once you zero the TX.

The step 3 result (41° instead of ~23°) is the part that seems wrong mathematically after zeroing TX the surface should land back at roughly the original trimmed angle. Would you be able to share which flight mode you were in during each step, and a full parameter dump? That would help narrow down where the discrepancy is coming from.

[orrerie]

Hello, thanks for a reply. I have done all of the abovementioned in Manual mode. It looks like a some sort of heisenbug to me

px4_trims.txt

[SofianElmotiem]

Hi again, looked at the parameter dump and found what's happening at step 3.

You have FW_ARSP_SCALE_EN = 1 and FW_USE_AIRSPD = 1. In FixedwingRateControl, TRIM_PITCH is applied multiplied by (FW_AIRSPD_TRIM / measured_airspeed)² to account for elevator effectiveness changing with airspeed. The stick input is not scaled this way. So on the ground, whatever your airspeed sensor is reading (even a small residual from wind or sensor noise), the trim gets amplified but the stick contribution doesn't.

At step 1 the elevator reads 23° because TRIM_PITCH is zero and only the stick matters. At step 3, TRIM_PITCH = 0.23 gets multiplied by that scaling factor. With an apparent airspeed of ~10 m/s on the ground, the factor is (14/10)² ≈ 1.96, giving roughly 0.23 × 1.96 = 0.45, which lands close to your 41°.

This is expected behavior. TRIM_PITCH is an aerodynamic trim meant to produce the right elevator moment at FW_AIRSPD_TRIM (14 m/s) in flight. The correct place to verify Copy Trims is in actual flight at trim airspeed, not on the ground. As a quick workaround for ground testing you can temporarily set FW_ARSP_SCALE_EN = 0, do the check, then re-enable it.

The step 2 transient (23° → 44°) is a separate thing: briefly both the live TX offset and the freshly written TRIM_PITCH are active at the same time. That resolves as soon as you zero the TX.