search

kriswiner / MPU9250

Arduino sketches for MPU9250 9DoF with AHRS sensor fusion
1.03k stars 472 forks source link

MPU9250 will be placed in random position - is there a way to get it working properly ? #281

Open jarekkt opened 6 years ago

jarekkt commented 6 years ago

Well, maybe not completely random, but still it will not be aligned to XYZ planes.

I have a device which my clients can attach to their boats in semi-random location and orientation. This means that I need to be ready to work even if e.g. the MPU 9250 XYZ planes are tilted. I can make factory calibration to fix base offsets ( for gyro/magnetometer/acceleremeter). User, after installation also makes single manual calibration ( so I know the installation orientation).

But then - what to do ? While gyro might not be affected, but magnetometer and acceleromter calibration does not seem to be trivial anymore. Especially that I cannot rotate the device anymore .

Has anyone developed runtime calibration for the MPU9250 ?

kriswiner commented 6 years ago

This is confusing.

Sensor fusion outputs quaternions from well-calibrated 9 DoFsensor data. The quaternions can be used to estimate absolute orientation of the device wrt gravity and magnetic north independent of any mounting considerations.

Are you not calibrating the sensors?

Are you not using sensor fusion to estimate absolute orientation?

On Mon, Jun 4, 2018 at 5:27 AM, jarekkt notifications@github.com wrote:

Well, maybe not completely random, but still it will not be aligned to XYZ planes.

I have a device which my clients can attach to their boats in semi-random location and orientation. This means that I need to be ready to work even if e.g. the MPU 9250 XYZ planes are tilted. I can make factory calibration to fix base offsets ( for gyro/magnetometer/ acceleremeter). User, after installation also makes single manual calibration ( so I know the installation orientation).

But then - what to do ? While gyro might not be affected, but magnetometer and acceleromter calibration does not seem to be trivial anymore. Especially that I cannot rotate the device anymore .

Has anyone developed runtime calibration for the MPU9250 ?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/281, or mute the thread https://github.com/notifications/unsubscribe-auth/AGY1qk--b-Wk3ygsf_Qr07FF18m1yOXtks5t5Se8gaJpZM4UY_jO .

jarekkt commented 6 years ago

Well, I do calibrate it - during production. Once.

But then the device is installed and its position is fixed in semi - random position. The position estimation work, is quite precise (both roll and pitch), having GPS also helps to adjust true heading.

I keep original calibration data in non-volatile memory, but it is not sufficient in long term. I can see that sensor heating takes its toll and it starts drifting. From tests you need about 5 minutes until it reaches stable point. I am also not quite sure about sensor aging - my device is rated for 10 years of lifetime and I have to be sure it is compensated

So I would like to recalibrate it. I cannot do it directly after power-on as it is not anymore in base position so I cannot simply take offsets directly from measurements.

I have asked about runtime calibration - DNP computing seems to detect 'no movement' and adjust zero offsets on the fly. In theory checking variations of measurements should detect standing still - especially for gyro. I cannot do much for magnetometer - during normal operation I can track measures, but mostly in two axes only ( my device is boat autopilot computer, should not be flipped :-) ) Accelerometer is also tricky, however I do have second source ( second redundant accelerometer, much more stable and precise )

Dnia 4 czerwca 2018 o 18:30 Kris Winer notifications@github.com napisał(a):

This is confusing.

Sensor fusion outputs quaternions from well-calibrated 9 DoFsensor data.
The quaternions can be used to estimate absolute orientation of the device
wrt gravity and magnetic north independent of any mounting considerations.

Are you not calibrating the sensors?

Are you not using sensor fusion to estimate absolute orientation?

On Mon, Jun 4, 2018 at 5:27 AM, jarekkt <notifications@github.com> wrote:

> Well, maybe not completely random, but still it will not be aligned to XYZ
> planes.
>
> I have a device which my clients can attach to their boats in semi-random
> location and orientation.
> This means that I need to be ready to work even if e.g. the MPU 9250 XYZ
> planes are tilted.
> I can make factory calibration to fix base offsets ( for gyro/magnetometer/
> acceleremeter).
> User, after installation also makes single manual calibration ( so I know
> the installation orientation).
>
> But then - what to do ? While gyro might not be affected, but magnetometer
> and acceleromter calibration does not seem to be trivial anymore.
> Especially that I cannot rotate the device anymore .
>
> Has anyone developed runtime calibration for the MPU9250 ?
>
> —
> You are receiving this because you are subscribed to this thread.
> Reply to this email directly, view it on GitHub
> <https://github.com/kriswiner/MPU9250/issues/281>, or mute the thread
> <https://github.com/notifications/unsubscribe-auth/AGY1qk--b-Wk3ygsf_Qr07FF18m1yOXtks5t5Se8gaJpZM4UY_jO>
> .
>

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/281#issuecomment-394417413 , or mute the thread https://github.com/notifications/unsubscribe-auth/AXzZ0YL_kSLAt_LYF98vMQpXKUMynXjQks5t5WCcgaJpZM4UY_jO .
kriswiner commented 6 years ago

If you are just using an MPU9250 and host-MCU-based fusion or simple tilt compensation then you will have to create the dynamic calibration algorithms yourself. You might take a look at the EM7180 motion co-processor which has a lot of this dynamic calibration built in.

https://www.tindie.com/products/onehorse/ultimate-sensor-fusion-solution/

On Mon, Jun 4, 2018 at 1:26 PM, jarekkt notifications@github.com wrote:

Well, I do calibrate it - during production. Once.

But then the device is installed and its position is fixed in semi - random position. The position estimation work, is quite precise (both roll and pitch), having GPS also helps to adjust true heading.

I keep original calibration data in non-volatile memory, but it is not sufficient in long term. I can see that sensor heating takes its toll and it starts drifting. From tests you need about 5 minutes until it reaches stable point. I am also not quite sure about sensor aging - my device is rated for 10 years of lifetime and I have to be sure it is compensated

So I would like to recalibrate it. I cannot do it directly after power-on as it is not anymore in base position so I cannot simply take offsets directly from measurements.

I have asked about runtime calibration - DNP computing seems to detect 'no movement' and adjust zero offsets on the fly. In theory checking variations of measurements should detect standing still - especially for gyro. I cannot do much for magnetometer - during normal operation I can track measures, but mostly in two axes only ( my device is boat autopilot computer, should not be flipped :-) ) Accelerometer is also tricky, however I do have second source ( second redundant accelerometer, much more stable and precise )

Dnia 4 czerwca 2018 o 18:30 Kris Winer notifications@github.com napisał(a):

This is confusing.

Sensor fusion outputs quaternions from well-calibrated 9 DoFsensor data. The quaternions can be used to estimate absolute orientation of the device wrt gravity and magnetic north independent of any mounting considerations.

Are you not calibrating the sensors?

Are you not using sensor fusion to estimate absolute orientation?

On Mon, Jun 4, 2018 at 5:27 AM, jarekkt notifications@github.com wrote:

Well, maybe not completely random, but still it will not be aligned to XYZ planes.

I have a device which my clients can attach to their boats in semi-random location and orientation. This means that I need to be ready to work even if e.g. the MPU 9250 XYZ planes are tilted. I can make factory calibration to fix base offsets ( for gyro/magnetometer/ acceleremeter). User, after installation also makes single manual calibration ( so I know the installation orientation).

But then - what to do ? While gyro might not be affected, but magnetometer and acceleromter calibration does not seem to be trivial anymore. Especially that I cannot rotate the device anymore .

Has anyone developed runtime calibration for the MPU9250 ?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/281, or mute the thread https://github.com/notifications/unsubscribe-auth/AGY1qk--b-Wk3ygsf_ Qr07FF18m1yOXtks5t5Se8gaJpZM4UY_jO .

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/281#issuecomment-394417413 , or mute the thread https://github.com/notifications/unsubscribe- auth/AXzZ0YL_kSLAt_LYF98vMQpXKUMynXjQks5t5WCcgaJpZM4UY_jO .

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/281#issuecomment-394487368, or mute the thread https://github.com/notifications/unsubscribe-auth/AGY1qrvRaLSfyp7yOq35nu0ij5PLEbbeks5t5ZfkgaJpZM4UY_jO .

jarekkt commented 6 years ago

Well, I was hoping for easy way - that somebody might have implemented it already.

It seems that I will have to do it myself, I will share results if I get to the point.

jarekkt commented 6 years ago

Well, I have checked what DMP algorithms do.

It seems that once it detects stable situation on accelerometers it just cancels gyro offsets.

kriswiner commented 6 years ago

DMP can only do 6 DoF fusion. If you are only worried about gyro offsets, then detecting no motion and doing gyro offsets on the host is easy to implement yourself.

On Tue, Jun 5, 2018 at 6:11 AM, jarekkt notifications@github.com wrote:

Well, I have checked what DMP algorithms do.

It seems that once it detects stable situation on accelerometers it just cancels gyro offsets.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/281#issuecomment-394703203, or mute the thread https://github.com/notifications/unsubscribe-auth/AGY1qvO9zBjf5_8GRww3-10r-HjvDQKEks5t5oOGgaJpZM4UY_jO .