PX4-Autopilot: Incorrect units of variance in GPS topic

Couple of issues and concerns which I had while upstreaming the GPS-VIO interface which will also allow an offboard estimator to use the GPS accuracy information:

1 . This is from the vehicle_gps_position topic :

float32 s_variance_m_s # GPS speed accuracy estimate, (metres/sec)
float32 c_variance_rad # GPS course accuracy estimate, (radians)

For velocity, variance should be in (m/s)^2 and for course, in rad^2. The comment and field name are both incorrect, and it is not understandable whether we want variance or standard deviation here.

1. From the mavlink message GPS_INPUT :

speed_accuracy float GPS speed accuracy in m/s
horiz_accuracy float GPS horizontal accuracy in m
vert_accuracy float GPS vertical accuracy in m

The units suggest that these are standard deviations, however field spec is not clear from the word “accuracy”. Judging from other messages with better docs e.g WIND_COV (horiz_accuracy float Horizontal speed 1-STD accuracy), it means sqrt variance.

1. Is there an objection to switching the vehicle_gps_position topic to covariances only (pos_cov and vel_cov), and derive the other uncertainty fields where needed? Seems kinda unnecessary to transport hdop, vdop, epv, eph, s_variance_m_s, c_variance_rad, etc. Also, modern estimators and any cascaded setup will benefit from the covariance availability. @LorenzMeier

I will send in a PR with the fixes and the new interface when as my concerns are confirmed by another pair of eyes.