Unexpected output of robot_localization EKF

edit

UPDATE I have changed the yaw_offset to 0 instead of 3.14. This has caused /odometry/filtered and odometry/gps to give much better results.

Setup

I have a TurtleBot3 robot with a GPS receiver. I am running ROS2 Dashing on Ubuntu 18.04 and a Raspberry Pi 3 Model B+. The transformation between /odom and /base_link is done in the TurtleBot3 bringup process. When I listen to the /odom everything seems good here. NB! The odometry topic has odometric data fused with IMU data. This is done in the hardware driver from ROBOTIS. In /odom: When I drive the robot forward, x-position increases. Moving to the left increases y-position. Turning counterclockwise around z gives an increase in yaw, so everything seems good here. I want to navigate outside so I want to use my GPS data for position information, and use the fused odometry and IMU data for orientation. I have set up an extended Kalman filter using the ekf_node from robot_localization and the navsat_transform node. The setup of these can be seen in the .yaml files below together with the launch file.

I have a tf2 transformation in the launch file since my GPS is mounted 23 centimeters above base_link and 8 centimeters behind. I have attached a picture (gps_transform) where this transform can be seen in rviz. It appears to be looking alright.

In the setup of the navsat_transform node I have a yaw_offset of 3.14.(Changed to 0 after update to post.) When I rotate my robot so that the imu_link x axis is pointing west I read 0 in yaw. From what the documentation says most IMUs have either north or east heading so there might be an issue here.

I have attached a ros2bag where the robot is stationary from 0:00 to 1:50 and starts moving thereafter named ekf_04_26-17_27 + a ros2bag with the new yaw_offset called ekf_28_24.

Questions

1. As the robot is moving forward the /odom is increasing in x as expected. However, /odometry/filtered is decreasing in y axis and the same goes for odometry/gps. How can that be? I suspect this may have something to do with the yaw_offset I have supplied to the navsat_transform node. UPDATE: This was caused by the wrong yaw_offset finding the correct yaw_offset solved this issue.

2. I can see that my /odometry/gps starts in position XYZ (0.08, 0.0, -0.23) from the transform I would expect the opposite XYZ (-0.08, 0.0, 0.23). Is something wrong with the transform itself or is it possible that the problem is arising from some other misconfiguration of the ekf setup? UPDATE: I am not sure that this is even a problem. It might be the correct transform. I just have a hard time grasping this one. Knowing that /odometry/gps is given in the map(world frame) and my transform is defined from /base_link to /gps my immediate expectation does not make sense.

3. At the beginning map, odom, base_link, imu_link and ...