I am currently trying to navigate using an RTK GPS. Unfortunately, despite calibration, my magnetometer's heading estimate seems to be slightly off at certain angles. While moving at these angles, this causes my robot to bank in one direction before turning back into the target (much like half of a heart shape). This makes sense as the robot is confident in the incorrect heading, drives in the wrong direction, and then spirals into target as the required heading drifts by more and more; if a set of waypoints are in a straight line, this issue can cause the robot to arrive at each waypoint perpendicular to the line.
This only occurs in a range of about 90 degrees (while functioning quite well in all other areas). Another separate issue is that the i2c bus will fail, requiring a power cycle. Any thoughts on that are also appreciated.
I have tried fusing multiple magnetometers (LIS3MDL and ICM-20948) together, but with only minor success.
I should also mention that my magnetometers were calibrated (in place, on the robot) with the following procedure: https://github.com/RigacciOrg/py-qmc5883l/tree/master/calibration, which does both hard and soft iron calibration. They are not tilt compensated, but I am operating them on pretty flat land.
I would like to avoid a dual antenna GPS solution due to cost and because I haven't found any documentation. I have also looked at using course-over-ground, but I am hoping that maybe there is a simpler solution that could get me up and running much sooner (there's also a lack of existing solutions for this).
Any help (big or small) would be much appreciated. Asides from fusing more magnetometers or spending a lot of time trying to implement course-over-ground, I'm not quite sure what my options are.
