robot_pose_ekf with only odometry information

edit

Hi rossers ;).

My robot delivers only very noisy odometry data. The worst character of that noise are sudden peaks of extremely unlikely data and sudden accelerations. I thought about using some kind of Bayesian filter to just filter that data based on an expected variance. Can I use the robot_pose_ekf-node for that? Of cause, it would not fusion any data. But can it just serve as a basic probabilistic filter? Will it also work on one data source to create a speed hypothesis?

Thanks for your help