Kiedro,
After a very long time with other matters I return to the Inertial sensor world of problems.
I read your threads again and I must say I'm amazed how much work and insights you have dug down into
this forum when it comes to sensor fusing and all math involved. I am truly impressed
I remember that I wrote a thread about Kalman filters some time ago
( http://www.mcselec.com/index2.php?option=com_forum&Itemid=59&page=viewtopic&t=11265&highlight=dummies )
and maybe it is time to contribute with something else soon...
Anyway, as some time has passed and new sensor devices has been introduced at an unstoppable pace.
I now wonder if someone here (perhaps you Kiedro) have evaluated the Invensense 9250 http://www.invensense.com/mems/gyro/mpu9250.html
It incorporate 3D Accel+Gyro plus input for 3D Magnetometers. And it also contains software for 9DOF sensor fusion.
All on the same chip (4x4mm) !!!
I have not yet read all the documentation, and I'm particularly interested to see if the firmware have support for Soft-iron calibration
of the magnetometers.
My current interest is to design a full 3D compass solution with hard and soft iron compensation ON THE FLY.
That is, adaptive recognition and elimination of offsets and non-linear flux for full 3D use.
The idea is that the device shall successively learn the distortion of the surrounding magnetic field lines due to magnetic objects attached to the
sensor frame, and eliminate it.
There are many many academic articles written about this problem and I have read most of them but still not found an engineering approach to the problem.
Many of them can do a static calibration with the use of massive matrix math, but few or none can do it as a natural part of the device operation.
The thing is that if you make a device that MUST be calibrated AFTER the device was mounted on the target equipment, then the procedure must
be simple enough for an unskilled person to perform it.
In practice, this doesn't work. End-users don't even want to get involved in these delicate matters. It should simply work out_of_the_box.
So the only solution for good sensor solutions today rely on two corner stones as i see it:
1: Multiple sensor fusion (and perhaps redundant sensors)
2. Adaptive calibration algorithms (Preferably no special modes. Just plain run)
So if you who bothered to read this far. Perhaps you could also give some thoughts or experience from how to turn academic
fire-flies into real embedded solutions...
Friday afternoon now. Time for a glass of wine and drift away from sensors and ugly software.
Cheers
/Per
Some links about magnetometer calibration:
----------------------------------------------
http://www.roboticsproceedings.org/rss09/p50.pdf
http://www.sgp.geodezja.org.pl/ptfit/wydawnictwa/krakow2011/APCRS%20vol.%2022%20pp.%209-23.pdf
http://www.ignss.org/LinkClick.aspx?fileticket=7oH5%2FnHXZyY%3D&tabid=88&mid=424
http://sailboatinstruments.blogspot.se/2011/08/improved-magnetometer-calibration.html <- I have tested this with good results, but it offers static calibration only
http://sailboatinstruments.blogspot.se/2011/09/improved-magnetometer-calibration-part.html
↧