Revision history [back]

Thanks for the pointers @Tom Moore, your suggestions helped narrow down the problem.

I got rid of the map frame EKF altogether to simplify things and used a single KF. Fused the absolute yaw from the IMU and yaw velocity from the wheel encoders. (We need an earth referenced heading in our system). This improved performance but the drift was not completely eliminated.

The culprit was the disagreement between the RTK poses and IMU orientations (Point 3 in @Tom's recommendations). With some tedious testing and adding a manual fudge factor to the IMU reported yaw we were able to get them aligned together.

For our long term solution, we procured a better IMU and with that have had superior performance.

Thanks for the pointers @Tom Moore, your suggestions helped narrow down the problem.

I got rid of the map frame EKF altogether to simplify things and used a single KF. Fused the absolute yaw from the IMU and yaw velocity from the wheel encoders. (We need an earth referenced heading in our system). This improved performance but the drift was not completely eliminated.

The culprit was the disagreement between the RTK poses and IMU orientations (Point 3 in @Tom's @Tom Moore's recommendations). With some tedious testing and adding a manual fudge factor to the IMU reported yaw we were able to get them aligned together.

For our long term solution, we procured a better IMU and with that have had superior performance.

Thanks for the pointers @Tom Moore, your suggestions helped narrow down the problem.

I got rid of the map frame EKF altogether to simplify things and used a single KF. Fused the absolute yaw from the IMU and yaw velocity from the wheel encoders. (We need an earth referenced heading in our system). This improved performance but the drift was not completely eliminated.

The culprit was the disagreement between the RTK poses and IMU orientations (Point 3 in @Tom Moore's recommendations). With some tedious testing and adding a manual fudge factor to the IMU reported yaw we were able to get them the data (LIDAR data and robot odometry) aligned together.

For our long term solution, we procured a better IMU and with that have had superior performance.

Thanks for the pointers @Tom Moore, your suggestions helped narrow down the problem.

I got rid of the map frame EKF altogether to simplify things and used a single KF. Fused the absolute yaw from the IMU and yaw velocity from the wheel encoders. (We need an earth referenced heading in our system). This improved performance but the drift was not completely eliminated.

The culprit was the disagreement between the RTK poses and IMU orientations (Point 3 in @Tom Moore's recommendations). With some tedious testing and adding a manual fudge factor to the IMU reported yaw we were able to get the data (LIDAR data and robot odometry) aligned together.

For our long term solution, we procured a better IMU and with that have had superior performance.performance since.