Hello!
I am trying to use the MPU 6000 in a project, and it seems to work well, but we have problems with the calibration.
If we put the eval board flat, the TILT angle (converted from the quaternions) is 0, but we need the 0 to be with the board perpendicular to this position. I have changed the orientation matrix, and I can see almost the correct TILT measurements. When I am in the zero position (with my "orientation", board perpendicular to floor) the TILT shows 5 degrees, or so.
So I started to look at the calibration process, and I found that in order to be able to calibrate with this new orientation, I have to substract the gravity from the correct axis that is perpendicular to earth. I have changed it, and it seems that the numbers are correct. The problem is that whatever position I set the calibration on, the TILT axis is never zero. I have followed the calibration functions, and everything seems to be fine (at the end, the gyro and accel bias are written to the DMP) but the output of the quaternions don't change very much, it seems
Any idea about how to achieve a 0º TILT in the correct position?
Thank you
Alex
- Log in to post comments
phpbb Topic ID
15948
Re: Re: MPU 6000 Accel calibration
I found that I had to calculate the acceleration biases as described in the Calibration Procedure but then subtract them, properly scaled to gs, from the acceleration data manually in the man program rather than using the acceleration bias registers in order to get the best results. I am trying to track this down; it might have to do with integer typing in my calibration script and/ or an incorrect masking of the temperature correction bits. Also, I had to use the sensitivity of the 16 g FSR in the bias registers not the 8 g FSR mentioned in the Calibration Procedure to get close. But in the best case using the registers I would often have an error of +/- 5 mg on at least one axis. I had no problem following the procedure with the gyros. That worked perfectly. The manual method always gave the most accurate results for acceleration biases. In this case, I could achieve consistent +/- 1 mg accuracy at rest by this method.