Revision history [back]

robot_pose_ekf, and kalman filters in general, are mostly used in robotics to provide improved odometry. They still drift a bit, but they're generally more accurate than any single relative motion sensor.

I don't think there's a way to force the EKF into a particular solution. Instead of forcing the EKF to take a particular value, most localization packages provide an additional transform from a global frame (usually /map) to the odometry frame (usually /odom or /odom_combined); this serves as a correction to the EKF data, and it can remain static if there are no localization updates (in your case, tag detections), while the EKF continues to publish relative position updates on top of the most recent localization result.