EKF velocity accumulation and attitude divergence after removing barometer and recompiling firmware (pure RTK, GPS-for-altitude configured)

I am using ArduPilot (EKF3) in a development project targeting a sealed / non-atmospheric environment, where the intention is to not use a barometer at all, and instead rely entirely on RTK (GPS) for both positioning and altitude estimation.

During testing, I removed the barometer backend at the board/source level, recompiled the firmware, and flashed it to the flight controller. As a result, no barometer driver exists in the system.

In addition, I have configured EKF3 strictly according to the official Wiki page “GPS for Altitude”
https://ardupilot.org/copter/docs/common-gps-for-alt
ensuring that altitude is sourced from GPS/RTK rather than a barometer.

However, despite this configuration, the following behavior is observed:

  • While the vehicle is stationary, ground speed continuously increases

  • EKF state gradually diverges over time

  • Attitude does not exhibit high-frequency jitter, but instead slowly drifts

  • The drift rate increases over time, eventually leading to large attitude errors and even roll/pitch inversion

  • GPS position itself appears reasonable, but the EKF-estimated velocity and attitude become inconsistent

From the observed behavior, this appears to be a case where vertical and/or velocity observability in the EKF becomes degraded after removing the barometer, causing the filter to rely more heavily on IMU integration, which then leads to velocity accumulation, position drift, and eventually attitude divergence.

I would appreciate clarification on the following points:

  1. In a pure RTK (GPS-only) use case, is it officially supported to fully remove the barometer at the source/firmware level?

  2. Even when EKF3 is configured per the GPS for Altitude Wiki, are there implicit EKF assumptions or dependencies on barometric altitude that remain?

  3. Has anyone encountered similar behavior where removing the barometer leads to continuous ground-speed growth and time-accelerating attitude divergence?

  4. For a barometer-free, pure-RTK setup, is there a recommended or validated EKF3 configuration or implementation path?

Any insights or guidance would be greatly appreciated. Thank you.

Can you post a log of an event where it is drifting?