Hello, I'm trying to implement robotlocalization in my robot, with IMU sensor which is mounted upside-down. I tried to change orientation of imuframe, but robotlocalization seems to ignore the changes. More precisely: it ignores orientation of imuframe when processing angular velocity (works correctly when processing orientation quaternion ~ when second line of imu0_config is "true, true, true" and fourth line is "false, false, false", but it does not fit my needs because of poor precision).

Details of my implementation:

robot_localization parameters are

odom_frame: odom
base_link_frame: base_footprint
world_frame: odom

two_d_mode: false

frequency: 50

odom0: husky_velocity_controller/odom
odom0_config: [false, false, false,
               false, false, false,
               true, false, false,
               false, false, false,
               false, false, false]
odom0_differential: false
odom0_queue_size: 10

imu0: imu/data
imu0_config: [false, false, false,
              true, true, false,
              false, false, false,
              false, false, true,
              true, false, false]
imu0_differential: false
imu0_queue_size: 10
imu0_remove_gravitational_acceleration: true

And I played the same bag file with topics needed by robotlocalization three times, each time using different orientation of imuframe.

case 1:

$ rosrun tf tf_echo base_footprint imu_link
At time 1537803279.035
- Translation: [-0.290, 0.000, 0.568]
- Rotation: in Quaternion [0.000, 0.000, 0.000, 1.000]
            in RPY (radian) [0.000, -0.000, 0.000]
            in RPY (degree) [0.000, -0.000, 0.000]

case 2:

$ rosrun tf tf_echo base_footprint imu_link
At time 1537803271.316
- Translation: [-0.290, 0.000, 0.568]
- Rotation: in Quaternion [1.000, 0.000, 0.000, -0.000]
            in RPY (radian) [-3.142, -0.000, 0.000]
            in RPY (degree) [-180.000, -0.000, 0.000]

case 3:

$ rosrun tf tf_echo base_footprint imu_link
At time 1537803264.615
- Translation: [-0.290, 0.000, 0.568]
- Rotation: in Quaternion [0.707, 0.000, 0.000, 0.707]
            in RPY (radian) [1.570, -0.000, 0.000]
            in RPY (degree) [89.954, -0.000, 0.000]

I expected that this experiment would result in three different trajectories of robot (in case 1 and case 2 robot should rotate in opposite directions and in case 3 there should have been no significant rotation at all, since the robot didn't rotate around other axes than z). But the reality is that the trajectories are exactly the same. For better explanation I append images of final robot position for each of the cases.

So is there a software solution for my problem? Is there parameter to enable rotated IMUs or is it a bug (or a reasoned limitation) of robotlocalization package? Or do I need to remount my IMU (which is, let's say, quite complicated)? If I do, does it need to be in the rotational center of robot (so that no centrifugal forces distort the measurements)? I thought that mounting IMU anywhere was supported by robotlocalization (also Documentation says it), but now I'm not completely sure about that... Thank you for your responses

Note: frameid of imu/data messages header is imudata and no misleading data (like odom->basefootprint tf, baselink->imu_data tf or /odometry/filtered) is included in the bag file.

Note 2: (if you want to reproduce my issue) for publishing imu_frame I use this node

Note 3: I use ros melodic and robot_localization version 2.5.2