Hello! Thanks for your excellent work.0
As is disscussed in many other issues, when using vins fusion with stereo+imu mode, we also suffers huge drift. After many experiments, we find that by setting sqrt_info matrix to be identical or divided by its norm, the drift seems disappeared. However, there seems a cordinate mismatch. For example, when we move the vehicle horizontally, the path of the algorithm indicates that vehicle drops first and rises when go back. The resulted path is below, in which the path should be at the same altitude.