hector_quadrotor_gazebo + imu_filter_madgwick, estimation quality?
Hello!
Does anyone have experience with the Madgwick
filter applied with the hector_quadrotor_gazebo
package (quadrotor_empty_world.launch
)?
I executed some maneuvers with the quadrotor in gazebo
and compared the published ground truth (/ground_truth_to_tf/pose
topic) orientation with the output of the Madgwick filter (/imu/data
topic). I observe some significant (too big) differences that are depicted in the following figures:
Roll - this is quite okey
Pitch - less okey, seems the drift is not compensated enough
Yaw - quite big differences, (soft/hard iron compensation, is it something that should be addressed here in gazebo?)...
I will start playing with the filter gain
which is now set to the default 0.1
. But i have a feeling that something is wrongly set up in my environment. So here are the relevant parts:
In spawn_quadrotor launch
file, i have:
<arg name="use_ground_truth_for_tf" default="true" />
<arg name="use_ground_truth_for_control" default="false" />
Once gazebo world with the quadrotor is initialized i start the imu_filter_madgwick
node in a launch file with the following params:
<param name="use_mag" value="true"/>
<param name="publish_tf" value="false"/>
<param name="use_magnetic_field_msg" value="false"/>
<param name="world_frame" value="enu"/>
<remap from="imu/data_raw" to="raw_imu"/>
<remap from="imu/mag" to="magnetic"/>
Basically that is it. While the maneuvers are executed i registered the data from /ground_truth_to_tf/pose
and /imu/data
topics to be able to make the above comparison.
Do you see something that is missing or wrongly set up? Any help is much appreciated! Thank you.
Asked by akosodry on 2019-09-26 06:19:37 UTC
Comments