small offsets in UKF odometry VS input odometry

[7DoF]

Hi, I'm running the UKF code with LIO odometry and raw IMU data. the thing is, everything is working well except, there are small offsets in UKF output odometry and my LIO odometry.

in this image: orange plot is LIO odometry (/Odometry/pose/pose/position/z) and pink plot is UKF output odometry plot(/odom_prop/pose/pose/position/z)

this offset constantly increases from takeoff and becomes constant later on (at a value of 15-23cm)

there is a similar thing happening in x direction as well but the offset value is just 8cm there, and there is almost no such offset in y direction.

I've tried changing the UKF config but there is not any visible change in offsets (max offset is still 23cm in z direction)

Any help is appreciated.

Thanks.

[fcladera]

Hi @7DoF ,

Did you verify your transformations between LiDAR and IMU? Both start at zero as this is the initialization value, but this may be a reason for the divergence.

[7DoF]

Hi @fcladera
about the extrinsic b/w Lidar and IMU:
the output fed into UKF is in IMU frame and the output of UKF is also in the same frame.
I have tried changing this extrinsic in the LIO code to see if the offset increases/decreases. but there is no diff in the offset.
(no offset in y, tiny offset in x, and about 15cm of offset in z]

I have been trying different things to understand the issue:

1. I ran ukf code in SITL setup with ground truth IMU data [noise free] and ground truth LIO Odometry data.
   • this is the graph and the scale of offset:

• as you can see the scale of offset is around 0.001m or .1cm here which is acceptable [ in z direction]
• another thing is, the scale of offset in x and y direction is around 0.001cm
• somehow this issue is more seen in z direction

2. another test that i did is:

• i started the UKF filter after takeoff
• my doubt is, that the IMU during takeoff is getting over exited somehow and because of this the UKF value is always more than the LIO odometry.
• when i start the UKF after takeoff, i notice that the offset are drastically reduced:
• here is an image: you can see that this time the offsets are very low.

can this be a filter thing that i can fix or this is just a raw sensor data issue? as in, the sensor outputs are offsetted during takeoff or during any jerky situation ?

thanks and regards

[fcladera]

What setup are you using? What is your IMU/LiDAR?

While I am not sure this may be related, vibration noise has always been a problem for us. We try to damp our sensors using these foam pads . You can also give it a try to this.

[7DoF]

Hi @fcladera, sorry for the late reply.

I've an ouster lidar and an external IMU.

I tried robot localization yesterday and these are the results I got:

The UKF here is following the input odometry really well throughout the flight. I'm getting a max offset of 1cm b/w UKF's propagated state and input odometry.

here is another result for drone takeoff type scenario:

As seen, there are no offsets that are getting added up here during takeoff.

This is the launch file i use for robot_localization UKF:

<launch>
<node pkg="tf2_ros" type="static_transform_publisher" name="static_transform_publisher1" args="0 0 0 0 0 0 base_link camera_init" />
<node pkg="tf2_ros" type="static_transform_publisher" name="static_transform_publisher2" args="0 0 0 0 0 0 camera_init map" />
<node pkg="tf2_ros" type="static_transform_publisher" name="static_transform_publisher3" args="0 0 0 0 0 0 odom camera_init" />

<node pkg="robot_localization" type="ukf_localization_node" name="ukf_localization" clear_params="true">
  <param name="frequency" value="30"/>  
  <param name="sensor_timeout" value="0.1"/>  
  <param name="two_d_mode" value="false"/>

  <param name="map_frame" value="map"/>
  <param name="odom_frame" value="camera_init"/>
  <param name="base_link_frame" value="base_link"/>
  <!-- <param name="world_frame" value="world"/> -->

  <param name="odom0" value="Odometry"/>
  <param name="imu0" value="/imu0"/> 


  <rosparam param="odom0_config">[true,  true,  true,
                                  true, true, true,
                                  true, true, true,
                                  false, false, false,
                                  false, false, false]</rosparam>

  <rosparam param= "imu0_config"> [false, false, false,
                                  true,  true,  true,
                                  false, false, false,
                                  true,  true,  true,
                                  true,  true,  true]</rosparam>

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

  <param name="imu0_remove_gravitational_acceleration" value="true"/>

  <param name="print_diagnostics" value="false"/>

  <!-- ======== ADVANCED PARAMETERS ======== -->

  <param name="odom0_queue_size" value="200"/>
  <param name="imu0_queue_size" value="1000"/>

  <param name="debug"           value="false"/>
  <param name="debug_out_file"  value="/home/tmoore/Desktop/debug_ekf_localization.txt"/>

  <rosparam param="process_noise_covariance">[0.05, 0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.05, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.06, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 0.03, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 0.0, 0.03, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 0.0, 0.0, 0.06, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.025, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.025, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.04, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 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, 0.0, 0.0, 0.0, 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, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.02, 0.0, 0.0, 0.0,
                                              0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 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, 0.0, 0.0, 0.0, 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, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.015]</rosparam>

       <rosparam param="initial_estimate_covariance">[1e-9, 0,    0,    0,    0,    0,    0,    0,    0,    0,     0,     0,     0,    0,    0,
                                                      0,    1e-9, 0,    0,    0,    0,    0,    0,    0,    0,     0,     0,     0,    0,    0,
                                                      0,    0,    0.1, 0,    0,    0,    0,    0,    0,    0,     0,     0,     0,    0,    0,
                                                      0,    0,    0,    1e-9, 0,    0,    0,    0,    0,    0,     0,     0,     0,    0,    0,
                                                      0,    0,    0,    0,    1e-9, 0,    0,    0,    0,    0,     0,     0,     0,    0,    0,
                                                      0,    0,    0,    0,    0,    1e-9, 0,    0,    0,    0,     0,     0,     0,    0,    0,
                                                      0,    0,    0,    0,    0,    0,    1e-9, 0,    0,    0,     0,     0,     0,    0,    0,
                                                      0,    0,    0,    0,    0,    0,    0,    1e-9, 0,    0,     0,     0,     0,    0,    0,
                                                      0,    0,    0,    0,    0,    0,    0,    0,    1e-9, 0,     0,     0,     0,    0,    0,
                                                      0,    0,    0,    0,    0,    0,    0,    0,    0,    1e-9,  0,     0,     0,    0,    0,
                                                      0,    0,    0,    0,    0,    0,    0,    0,    0,    0,     1e-9,  0,     0,    0,    0,
                                                      0,    0,    0,    0,    0,    0,    0,    0,    0,    0,     0,     1e-9,  0,    0,    0,
                                                      0,    0,    0,    0,    0,    0,    0,    0,    0,    0,     0,     0,     1e-9, 0,    0,
                                                      0,    0,    0,    0,    0,    0,    0,    0,    0,    0,     0,     0,     0,    1e-9, 0,
                                                      0,    0,    0,    0,    0,    0,    0,    0,    0,    0,     0,     0,     0,    0,    1e-9]</rosparam>

</node>
</launch>

I still haven't gone through the code base for robot localization. but my understanding is state propagation in FLAUKF is left incomplete maybe? in the sense that the covariance scales are mismatched and because of this propagation is only happening in an an obvious sort of manner. (I mean the tiny corrections which maybe in the order of a 1cm or 2cm that have to be appended to the state are hard to estimate)

the thing is I've got most offset issues in Z direction, which i think is one of the hardest state to maintain because the drone has to counter gravity and then maintain its state.

Anyway, this is just my understanding. Let me know what you think about this.

Thanks