Revision history [back]

I'm looking at your config file, and I have a few comments:

1. Are you trying to run all of these at the same time? That's going to cause all sorts of trouble in rviz, because they are all publishing the same transform. If you want to do an apples-to-apples comparison, I suggest you rostopic echo /some_filtered_topic -p > output.txt, and then plot the output using MATLAB or Octave.
2. I'll tell you what I tell a lot of people: start simple. Forget the GPS for a moment, and forget testing all the models at once. Start with one simple EKF config that just fuses IMU and wheel encoder velocities from one of your models. Get that working well, and then plot the data. Then try with a different model, and plot that on top of it.
3. You have two_d_mode turn on, but you are fusing roll and pitch in your IMU. It won't hurt anything, but it also won't actually do anything.
4. Your first data sources are not fusing yaw velocity at all. I assume that's intentional.
5. For wheel encoder data, you should always fuse the 0 value for y velocity (just set Y velocity to true). Give it a tiny covariance value, too. Your robot is not omnidirectional, so fusing a 0 value for Y enforces a kinematic constraint that stops the robot from moving sideways.