I'm trying to make INS with MPU9250.
I followed the steps below.
- calculate the horizontal acceleration with roll, pitch, yaw and ax, ay, az
- use the average filter to make the current acc (0, 0, 0) (just when the measurement starts)
I measure 2000 times to calculate the average of horizontal acceleration
- with the above acc, calculate location
Then, it works when it is at the initial posture but when I lean over the sensor, the horizontal acceleration doesn't result (0, 0, 0), but there is an error less than plus or minus 2, which belongs to a big error to location.
I want to improve my INS result. Is it a limit of the sensor? Is there any mathematical method to reduce the error?
Please help me...