Problem when fusing IMU and Odometry

edit

Hi, I am trying to use robot localization to fuse the data from an XSense IMU and Odometry Source (Pioneer). The output is not what i expect actually. The resulting odometry looks similar (almost exact) to the odometry of the pioneer? So the filter is not improving or at least changing the result?

This is my launch file:

<param name="frequency" value="30"/>
<param name="sensor_timeout" value="0.1"/>
<param name="base_link_frame" value="base_footprint"/>
<param name="two_d_mode" value="true"/>

<param name="imu0" value="imu/data"/>
<rosparam param="imu0_config">[false, false, false,
                           false, false,  true,
                           false, false, false,
                           false, false, false,
                           false,  false, false]</rosparam>
<param name="imu0_differential" value="true"/>
<param name="imu0_remove_gravitational_acceleration" value="false"/>

<param name="odom0" value="/odom/fixed_wheel"/>
<rosparam param="odom0_config">[true, true,  false, 
                            false, false, true, 
                            false, false, false, 
                            false, false, false,
                            false, false, false]</rosparam>

  <param name="odom0_relative" value="true"/>
  <param name="imu0_relative" value="true"/>

Here is an sample bagfile together with robot localization: https://www.dropbox.com/s/kalrug9s48j5w3x/odom_imu_fuse.bag?dl=0

Topic: /imu/data == xsense imu data
Topic: /odom/fixed_wheel == pioneer improved odometry (by hand added covariances)
Topic: /odom/ukf == current results of robot localization
Topic: /scan/sick == laser scanner for reference
Topic: /tf
Topic: /tf_static

Here is the same bagfile without robot localization: https://www.dropbox.com/s/01xtvgmv7gchb63/odom_imu_unfused.bag?dl=0 --- Update ---

I added this fixed odometry covariance:

odom.pose.covariance =  boost::assign::list_of(1e-2) (0) (0)  (0)  (0)  (0)
                                             (0) (1e-2)  (0)  (0)  (0)  (0)
                                             (0)   (0)  (1e6) (0)  (0)  (0)
                                             (0)   (0)   (0) (1e6) (0)  (0)
                                             (0)   (0)   (0)  (0) (1e6) (0)
                                             (0)   (0)   (0)  (0)  (0)  (1e-3) ;

Maybe i am missing something obvious? Thanks in advance!

--- Update ---

I tried some different covariance matrices but nothing changed but when i changed my launch file like so:

    <param name="imu0_differential" value="true"/>
    <param name="odom0_differential" value="true"/>

then i got following result: image its an image, green is the wheel odometry and red the ukf result and blue is the ground truth (i know, really bad drawn^^)

What i do not understand is why the imu is making the result worse? This data is captures outside on a sidewalk, could this be a problem? I am pretty sure that the imu is directed in the right direction but it looks weird...

-- Update --

Here is the sample output of each source, first odometry, then imu:

header: 
 seq: 4343
 stamp: 
  secs: 1464844455
  nsecs: 543424495
 frame_id: odom
 child_frame_id: ''
pose: 
 pose: 
 position: 
  x: -68.2792701104
  y: 122.427959131
  z: 0.0
 orientation: 
  x: 0.0
  y: -0.0
  z: -0.629156547298
  w: -0.777278610919
 covariance: [0.01, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.01, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1000000.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1000000.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1000000.0, 0.0, 0.0, 0.0, 0.0, 0.0 ...