Revision history [back]

Thank you for including sample sensor data. Your config looks OK to me. Can you visualize the poses that are being generated by the Hector node? If the EKF generally tracks those poses, then you can try inflating the Hector covariances. Your process_noise_covariance is already pretty small, and the integrated error from the twist data ought to also be pretty tiny, so I would expect the filter to trust your odometry data.

Also, what happens if you just don't include the Hector data, and fuse only wheel encoder data? Does it stay still?