Revision history [back]

Check out the sensor_msgs/NavSatFix documentation here.

To cite:

# Position covariance [m^2] defined relative to a tangential plane
# through the reported position. The components are East, North, and
# Up (ENU), in row-major order.

and

# If the covariance of the fix is known, fill it in completely. If the
# GPS receiver provides the variance of each measurement, put them
# along the diagonal. If only Dilution of Precision is available,
# estimate an approximate covariance from that.

About where they come from: Which driver are you using? The nmea_navsat driver approximates the covariances from HDOP values (see the respective lines here)

You cannot directly get better position estimates using the covariances, but you know about the error that exists. And the covariances can be used in sensorfusion to get a better overall estimate.