search

JokerJohn / LIO_SAM_6AXIS

LIO_SAM for 6-axis IMU and GNSS.
628 stars 123 forks source link

IMU preintegration jerky #25

Closed gwl27 closed 2 years ago

gwl27 commented 2 years ago

Dear John,

great work with that repo!

If I test it with the provided datasets everything works as expected.

Now I have a custom dataset, where the tf are a bit different. The LiDAR odometry works still although the IMU preintegration seems to be off by a fair bit as show in the screenshot. Do you have any clue what could cause that and what I could check?

The parameters extrinsicRot and extrinsicRPY seem ok. Could this also happen because of wrong noise parameters or must there be something of in the tf tree?

Thanks a lot, gwl27 Screenshot from 2022-09-28 15-55-41

JokerJohn commented 2 years ago

Hi, @gwl27, Thanks for your feedback！ From your picture, the extrinsics extrinsicRot and extrinsicRPY settings should be right. One question is, the motion of the sensor and your sensor platform, whether robust or relatively stable, handheld or vehicle platform. At the beginning of your dataset, It should be normal to use a handheld platform to do aggressive motion since there is a large gap between the pose predicted by the IMU preintegration and the pose obtained by point cloud matching. Also, if your IMU is consumer-grade and low-cost, this phenomenon is in the normal category.