Calibration - figure of eight

[IanBUK]

I'm being a bit thick. My mpu9250 sensors look to be baldy calibrated and I'm wondering what I've done wrong. I read about the 'figure of eight' and am thinking 'wait....'. So silly question time: when moving the sensor in a figure of eight, do I keep it heading in the same direction keeping the yaw constant, or should I rotate it in the direction of travel, rotating through yaw? Should I be moving it flat on my desk, or vertically? Ok, that's two questions.

I might as well ask a third....when I flip the sensor through any axis, the reading take a long time to catch up. As in tens of seconds, not tenths of seconds.

Ok, a fourth, and final question. If I leave the sensor alone, I'm seeing pitch drift over time. Why might this be?

I'd really appreciate some pointers on this, thanks.

Ian.

[kriswiner]

So these problems are likely due to incomplete calibration and selecting the incorrect reference frame. For accel and gyro calibration, leaving the sensor flat and motionless on a table top should be good enough fr reasonable accuracy. For the mag, the idea is to sample the entire spherical response surface multiple times during calibration. So imagine an arrow pointing up from the MPU9250 pckage. The goal of mag calibration is to point this arrow in every possible direction (i.e., every point on the inner surface of a big balloon surrounding the sensor) multiple times. This usually takes a minute or two. Then you need to pass the sensor data to the Madgwick or Mhony filter in NED coordinates, meaning aN, aE, aD, gN, gE, gD, mN, mE, mD order (at least for Madgwick). Could also use ENU I suppose. So the idea is to pick which board direction you want to point to absolute North. Then identify which sensor axes are aligned in this direction, there are then the N values. Then once you know this, which are the E and D values? Pass these into the fusion function and you should get resonable results. The last part of the recipe is to use an MCU with enough power to sun the fusion filter 10 - 20 times for each new gyro value. So if you run the gyro at 200 Hz, run the fusion filter at 2 kHz or 4 kHz, etc.

…

On Tue, Sep 5, 2023 at 2:41 PM IanBUK ***@***.***> wrote: I'm being a bit thick. My mpu9250 sensors look to be baldy calibrated and I'm wondering what I've done wrong. I read about the 'figure of eight' and am thinking 'wait....'. So silly question time: when moving the sensor in a figure of eight, do I keep it heading in the same direction keeping the yaw constant, or should I rotate it in the direction of travel, rotating through yaw? Should I be moving it flat on my desk, or vertically? Ok, that's two questions. I might as well ask a third....when I flip the sensor through any axis, the reading take a *long* time to catch up. As in tens of seconds, not tenths of seconds. Ok, a fourth, and final question. If I leave the sensor alone, I'm seeing pitch drift over time. Why might this be? I'd really appreciate some pointers on this, thanks. Ian. — Reply to this email directly, view it on GitHub <#491>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABTDLKTHIVLPQNR5UFZLXFLXY6L7DANCNFSM6AAAAAA4MMJFMI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[IanBUK]

Thanks for the quick response. I shall have a go over the next few days, see where I get to and update here.Thanks,Ian.On 5 Sep 2023, at 23:04, Kris Winer ***@***.***> wrote: So these problems are likely due to incomplete calibration and selecting the incorrect reference frame. For accel and gyro calibration, leaving the sensor flat and motionless on a table top should be good enough fr reasonable accuracy. For the mag, the idea is to sample the entire spherical response surface multiple times during calibration. So imagine an arrow pointing up from the MPU9250 pckage. The goal of mag calibration is to point this arrow in every possible direction (i.e., every point on the inner surface of a big balloon surrounding the sensor) multiple times. This usually takes a minute or two. Then you need to pass the sensor data to the Madgwick or Mhony filter in NED coordinates, meaning aN, aE, aD, gN, gE, gD, mN, mE, mD order (at least for Madgwick). Could also use ENU I suppose. So the idea is to pick which board direction you want to point to absolute North. Then identify which sensor axes are aligned in this direction, there are then the N values. Then once you know this, which are the E and D values? Pass these into the fusion function and you should get resonable results. The last part of the recipe is to use an MCU with enough power to sun the fusion filter 10 - 20 times for each new gyro value. So if you run the gyro at 200 Hz, run the fusion filter at 2 kHz or 4 kHz, etc.

On Tue, Sep 5, 2023 at 2:41 PM IanBUK ***@***.***> wrote: I'm being a bit thick. My mpu9250 sensors look to be baldy calibrated and I'm wondering what I've done wrong. I read about the 'figure of eight' and am thinking 'wait....'. So silly question time: when moving the sensor in a figure of eight, do I keep it heading in the same direction keeping the yaw constant, or should I rotate it in the direction of travel, rotating through yaw? Should I be moving it flat on my desk, or vertically? Ok, that's two questions. I might as well ask a third....when I flip the sensor through any axis, the reading take a *long* time to catch up. As in tens of seconds, not tenths of seconds. Ok, a fourth, and final question. If I leave the sensor alone, I'm seeing pitch drift over time. Why might this be? I'd really appreciate some pointers on this, thanks. Ian. — Reply to this email directly, view it on GitHub <#491>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABTDLKTHIVLPQNR5UFZLXFLXY6L7DANCNFSM6AAAAAA4MMJFMI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>