Chao Qu

Do you think these jumps are timestamped incorrectly (by the driver) or are they output incorrectly by the sensor?

I wouldn't say it's "infinitely high", judging from the plot here unless you are fixing the axis limits then the drop-offs are a fixed +-0.004s. What I suggested before is...

@IanTheEngineer Thanks for testing, do you think you can pull the fix from the other branch to master?

I can also verify the fix. Plot for delta time Before:  After:  I will open a PR soon with the fix I pulled in from dashing.

There might be one tiny problem with this fix, which is the time stamp will slowly deviate from host time. This plot shows ros::time::now() - fixed_time in ns 

I think point 2 might be the reason. Have you tried this one https://github.com/ros-drivers/microstrain_mips This supports gx5 while ours has not been tested with one.

There could be many reasons that cause messages to drop. (maybe system is under heavy load). If only a few dropped from 4000 msgs then I wouln't worry too much.

The driver does support filtered orientation. One thing you could try is to rotate the acceleration into the world frame given the filtered orientation and subtract the gravity and rotate...

Other imu products usually output magnetic field in uT https://www.xsens.com/wp-content/uploads/2015/05/MTi-100-series.pdf http://www.chrobotics.com/docs/UM7_Datasheet.pdf

Sounds good. I will change it to Tesla then.