New accuracy analysis using pyulog (px4tools seems to no longer be maintained). Some new insight into how flight altitude affects accuracy of estimates.
As can be seen from the histograms - estimates are all biased, some potential causes:
Poor test design: drone and camera facing only east with slight perspective in camera causes bias in easting/northing directions
Comparing AGL altitude against AMSL altitude causes bias in altitude/z-axis direction as simulation starting location is a few meters above sea level
Potential undiscovered issues with implementation, esp. coordinate frame conversions
New accuracy analysis using pyulog (px4tools seems to no longer be maintained). Some new insight into how flight altitude affects accuracy of estimates.
As can be seen from the histograms - estimates are all biased, some potential causes: