Revision history [back]

robot_localization provides a nicely configurable Kalman Filter implementation and is probably the most frequently used package for such tasks. That being said, if you only have wheel odometry available do not expect major improvements, as KF-based estimators are most useful for fusing multiple different sources of information. A common setup would be fusing angular yaw rates from an IMU with linear velocities from wheel odometry.

If you just send wheel odometry to a EKF and nothing else. By adjusting the process and measurement noise you could add some smoothing/adjust for noise, but to see noticable improvements adding for instance an IMU to measure angular rates is recommended.