How to find reasonable values for NavSatFix position covariance?

I have a GPS sensor whose driver doesn't report any info on positioning accuracy. Basically, what we have is lon, lat, alt only.

Is there a way to make up reasonable values for the position covariance in order to feed the data into a Kalman filter? (I assume that robot_localization's EKF/UKF doesn't insert useful default values in case the covariance is reported all zero...)

I'd be interested in any insights, from ballpark estimates to calibration methods or anything else.

Asked by moooeeeep on 2019-11-22 04:54:54 UTC

Comments

Answers

You could estimate the covariance matrices using trial and error. If not outrageously wrong, the covariances will converge eventually if my memory serves me right. Another method would be to apply the Autocovariance Least-Squares (ALS) technique (or possible other techniques) to estimate the noise covariances directly from sampled data. A quick search resulted in this paper which talks about the technique: Paper Link

Asked by davekroezen on 2019-11-22 05:43:10 UTC