[EKF2] Tilt estimate improvements

[bresch]

Solved Problem

Constant small accelerations during gravity fusion creates a small tilt error. As a consequence, the accel bias gets incorrectly estimated and the gravity fusion (using this bias estimate) takes more time to correct back the error.

Solution

Stop accel bias estimation while gravity fusion is active.

Additionally:

• start gravity fusion based on accelerometer LPF instead of a peak hold filter (this was causing the flag to switch on and off all the time due to vibrations/spikes). The innovation filter with a small gate (0.25) is more than enough to avoid fusing quick changes in accelerometer data
• add a "update_tilt" parameter to declination to force the "tilt Kalman gains" to 0 as we already do in mag fusion

Test coverage

EKF replay, unit tests, SITL tests

Before: rapid flag switching

After: incorrect acceleration are rejected by he innovation filter and not the control flag

[dagar]

_accel_lpf is only being updated during init, but I was thinking we should move it regardless.

PX4-Autopilot/src/modules/ekf2/EKF/ekf.cpp

Lines 180 to 198 in beb7e7d

	bool Ekf::initialiseFilter()
	{
	// Filter accel for tilt initialization
	const imuSample &imu_init = _imu_buffer.get_newest();
	// protect against zero data
	if (imu_init.delta_vel_dt < 1e-4f || imu_init.delta_ang_dt < 1e-4f) {
	return false;
	}
	if (_is_first_imu_sample) {
	_accel_lpf.reset(imu_init.delta_vel / imu_init.delta_vel_dt);
	_gyro_lpf.reset(imu_init.delta_ang / imu_init.delta_ang_dt);
	_is_first_imu_sample = false;
	} else {
	_accel_lpf.update(imu_init.delta_vel / imu_init.delta_vel_dt);
	_gyro_lpf.update(imu_init.delta_ang / imu_init.delta_ang_dt);
	}

[dagar]

I was moving it here, but got a little sidetracked. #24046

[bresch]

Ouch, I didn't realize that; thanks