Spatial precision - underreported velocity about phidgets_drivers

[mintar]

According to the second plot, the two IMUs disagree by about 0.02 rad/s (approx. 1°/s), so that doesn't sound too bad. I believe that shouldn't lead to the problems you're experiencing. You can test that theory by using the T265's IMU as input to your EKF; if the problem persists, you can rule out the velocity difference between the IMUs as an error source.

Another thing to note is that I see this bot with the /imu/data_raw and /imu/data coming out from the madgwick filter.

Nope, the madgwick filter subscribes to /imu/data_raw, fills the orientation field and republishes the data to /imu/data.

In summary, I suspect that there's something wrong with your EKF configuration.

[msadowski]

Thanks for a swift reply @mintar! Indeed you are correct that my problem will lie somewhere else. Here is the same setup using EKF fusing the Realsense IMU:

The first thing that strikes me is that now the odometry coming out of EKF is rotated to the right w.r.t. the pure wheel odom (with the Phidget it was tilted to the left). It might be somehow related to the fact that the realsense is mounted in the front of the robot and Phidget on the back.

I'll close this issue now as it's clearly not the Phidget's issue. If you happen to have any EKF advice then I'd happily take it!