Analyzing localization on real robot

edit

I'm looking for an effective and reliable way to test the localization system (amcl) on my Pioneer LX robot. My idea is to plot the x, y, and theta covariances for different amcl parameters when driving the Pioneer LX from the same start position to the same end position. However, I'm not sure what is the best way to do this in order to eliminate other random factors (such as differences in start and end positions). Would manually placing the robot on the start position and then record a .bag file be sufficient, or are there better methods for this?