fix(ekf2): hard reset external position estimate while position aiding active

[haumarco]

Solved Problem

Since #25223, MAV_CMD_EXTERNAL_POSITION_ESTIMATE is also executed while a horizontal position source (GNSS, EV, AGP, beacon) is being fused. In that case the command takes the reset-by-fusion path introduced in #25885, which applies the innovation with an artificially high Kalman gain and deliberately pushes it into the correlated states.

This is correct during dead reckoning (large position variance, innovation reflects accumulated state error), but harmful with an active position source: the position variance is small, so the gain into correlated states is large, and the innovation is typically caused by a bias of the aiding source rather than by state error. In a real flight this yanked the velocity state by ~18 m/s from an 83 m position innovation, which then caused airspeed and position fusion to be gate-rejected.

Solution

Route the command to the hard-reset branch whenever isHorizontalPositionAidingActive() is true. The hard reset translates the position, keeps velocity (minus wind), and zeroes wind with inflated variance. Reset-by-fusion remains in place for its intended dead-reckoning use case.

Changelog Entry

For release notes:

Bugfix: MAV_CMD_EXTERNAL_POSITION_ESTIMATE now performs a hard position reset instead of reset-by-fusion while a horizontal position aiding source (GNSS, EV, AGP, beacon) is active, preventing large velocity/wind/heading corruption from the forced fusion

Test coverage

Tested in SITL

[github-actions]

🔎 FLASH Analysis

px4_fmu-v5x [Total VM Diff: 32 byte (0 %)]

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.0%     +32  +0.0%     +32    .text
    +2.1%     +28  +2.1%     +28    Ekf::resetGlobalPosToExternalObservation()
    +0.2%      +4  +0.2%      +4    Ekf::fuseMag()
  +0.0%     +55  [ = ]       0    .debug_abbrev
  +0.0%     +92  [ = ]       0    .debug_info
  +0.0%     +53  [ = ]       0    .debug_line
     +50%      +1  [ = ]       0    [Unmapped]
    +0.0%     +52  [ = ]       0    [section .debug_line]
  +0.0%     +78  [ = ]       0    .debug_loclists
  +0.0%      +2  [ = ]       0    .debug_rnglists
    [NEW]      +1  [ = ]       0    [Unmapped]
    +0.0%      +1  [ = ]       0    [section .debug_rnglists]
  -0.4%     -32  [ = ]       0    [Unmapped]
  +0.0%    +280  +0.0%     +32    TOTAL

px4_fmu-v6x [Total VM Diff: 32 byte (0 %)]

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.0%     +32  +0.0%     +32    .text
    +2.1%     +28  +2.1%     +28    Ekf::resetGlobalPosToExternalObservation()
    +0.2%      +4  +0.2%      +4    Ekf::fuseMag()
  +0.0%     +55  [ = ]       0    .debug_abbrev
  +0.0%     +92  [ = ]       0    .debug_info
  +0.0%     +53  [ = ]       0    .debug_line
   -50.0%      -3  [ = ]       0    [Unmapped]
    +0.0%     +56  [ = ]       0    [section .debug_line]
  +0.0%     +20  [ = ]       0    .debug_loclists
  -0.6%     -32  [ = ]       0    [Unmapped]
  +0.0%    +220  +0.0%     +32    TOTAL

Updated: 2026-07-10T07:00:40

[bresch]

So external pos aiding becomes external pos reset if another source of position is active; si it gets the absolute priority over everything else. I guess that's fin as long as we only use it for manual position updates (i.e.: only a couple of updates, not an external sensor with a constant update rate)