3d robot_localization and bounding Yaw
I'm fusing the wheel odom and imu on the Husky, with robot_localization's UKF. What I'm wondering is the following:
I don't really want to rely on the magnetometer output for bounding yaw from the IMU, but I also don't want to orient my yaw around my wheel encoder yaw. So, I'd like to just fuse both as angular velocities, and let it slowly drift with gyro bias.
However, when I do this (fuse YawRate from odom, YawRate from IMU) the covariance for YawYaw (/odometry/filtered/pose/covariance[35]) grows slowly. After a few hours, it will be very large.
Is this a significant issue? I assume the filter will become unstable (I'm using the UKF currently). Thus, do I have to fuse an absolute yaw from somewhere in order to keep the filter stable?
As a side note, the covariance for ZZ is also growing (/odometry/filtered/pose/covariance[14]), but I don't think that matters as much.
