Computing Beta and Zeta Values

[mecitpmk]

Hello , I was trying to tune the beta and zeta parameters to make the orientation better.

I was came across with those parameeters in the codes :

float GyroMeasError = pi * (40.0f / 180.0f);   // gyroscope measurement error in rads/s (start at 40 deg/s)
float GyroMeasDrift = pi * (0.0f  / 180.0f);   // gyroscope measurement drift in rad/s/s (start at 0.0 deg/s/s)

How did you determine the gyroscope measurement error as 40 deg/s ? I couldnt see anything about it in the datasheet. Besides that , same question for the gyroscope measurement drift as 0.0 deg/s/s .

If you have any knowledge about it please inform me about that.

Thank you,

[kriswiner]

These parameters are irrelevant to orientation quality. Better to focus on good calibration of the sensors, especially mag.

On Mon, Dec 12, 2022 at 1:39 PM mecit @.***> wrote:

Hello , I was trying to tune the beta and zeta parameters to make the orientation better.

I was came across with those parameeters in the codes :

float GyroMeasError = pi (40.0f / 180.0f); // gyroscope measurement error in rads/s (start at 40 deg/s) float GyroMeasDrift = pi (0.0f / 180.0f); // gyroscope measurement drift in rad/s/s (start at 0.0 deg/s/s)

How did you determine the gyroscope measurement error as 40 deg/s ? I couldnt see anything about it in the datasheet. Besides that , same question for the gyroscope measurement drift as 0.0 deg/s/s .

If you have any knowledge about it please inform me about that.

Thank you,

— Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/485, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTDLKWG47FIMNNP3RMQXHLWM6LP5ANCNFSM6AAAAAAS4OAA7Y . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[mecitpmk]

Thank you for the fast reply @kriswiner .

You are right, I should say the response time not the orientation quality. Should I set those parameters with tuning and observing method to get better response time or its related with the datasheet informations?

Thank you,

[kriswiner]

These are fusion filter parameters, and have nothing to do with the underlying sensor properties. You could spend some time trying to adjust these parameters but it is likely a waste of time. Better, after achieving good calibration, to use a fast MCU. Something like a CM4F or Teemsy 3.2 or 4.0 for example. Even an ESP32 would work well. This is because the Madgwick filter is iterative and one should run the fusion filter 10 - 20 times for every gyro data sample, and one should run the gyro at 200 - 800 Hz. So slow MCUs will give very high latency....this is much more important than these filter parameters....

On Mon, Dec 12, 2022 at 2:01 PM mecit @.***> wrote:

Thank you for the fast reply @kriswiner https://github.com/kriswiner .

You are right, I should say the response time not the orientation quality. Should I set those parameters with tuning and observing method to get better response time or its related with the datasheet informations?

Thank you,

— Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/485#issuecomment-1347400009, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTDLKTQD4QNDSOAQ4UDE3DWM6OCZANCNFSM6AAAAAAS4OAA7Y . You are receiving this because you were mentioned.Message ID: @.***>

[mecitpmk]

Dear Winer,

Thank you for the repyling the content.

I was researching and going deeply into this Madgwick filter and I research iteration about Madgwick filter and I found this :

According to Magdwick : "A conventional approach to optimisation would require multiple iterations of equation (19) to be computed for each new orientation and corresponding senor measurements. Efficient algorithms would also require the step-size µ to be adjusted each iteration to an optimal value; usually obtained based on the second derivative of the objective function, the Hessian. However, these requirements considerably increase the computational load of the algorithm and are not necessary in this application. It is acceptable to compute one iteration per time sample provided that the convergence rate governed by µt is equal or greater than the physical rate of change of orientation."

So I am confused about that. Do we really need to compute 10-20 times?

@kriswiner

Thank you,

[kriswiner]

yes, but try yourself to confirm

On Mon, Dec 12, 2022 at 2:24 PM mecit @.***> wrote:

Dear Winer,

Thank you for the repyling the content.

I was researching and going deeply into this Madgwick filter and I research iteration about Madgwick filter and I found this :

According to Magdwick : "A conventional approach to optimisation would require multiple iterations of equation (19) to be computed for each new orientation and corresponding senor measurements. Efficient algorithms would also require the step-size µ to be adjusted each iteration to an optimal value; usually obtained based on the second derivative of the objective function, the Hessian. However, these requirements considerably increase the computational load of the algorithm and are not necessary in this application. It is acceptable to compute one iteration per time sample provided that the convergence rate governed by µt is equal or greater than the physical rate of change of orientation."

So I am confused about that. Do we really need to compute 10-20 times?

Thank you,

— Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/485#issuecomment-1347427674, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTDLKSCFMDPC7WXZ4JZYJTWM6Q25ANCNFSM6AAAAAAS4OAA7Y . You are receiving this because you were mentioned.Message ID: @.***>

[mecitpmk]

I will try and share the result.

Thank you,

[mecitpmk]

I tried and come up with those result :

Test 1: Assume that the sensor frequency = 1000Hz, Asseme that the Madgwick filter freq = 1000Hz, Result : Orientation filter response is much more clear and faster.

Test 2: Assume that the sensor frequency = 1000Hz, Asseme that the Madgwick filter freq = 2000Hz, Result : Orientation filter response is much more slower and delayed.

[kriswiner]

Assume? Do you mean measure?

On Sat, Dec 17, 2022 at 4:48 PM mecit @.***> wrote:

I tried and come up with those result :

Test 1: Assume that the sensor frequency = 1000Hz, Asseme that the Madgwick filter freq = 1000Hz, Result : Orientation filter response is much more clear and faster.

Test 2: Assume that the sensor frequency = 1000Hz, Asseme that the Madgwick filter freq = 2000Hz, Result : Orientation filter response is much more slower and delayed.

— Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/485#issuecomment-1356594384, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTDLKQVRTAY2ZWZS3ARM7TWNZNPVANCNFSM6AAAAAAS4OAA7Y . You are receiving this because you were mentioned.Message ID: @.***>

[mecitpmk]

Yes, the experiment results based on the real IMU measurements.

According to my application, ( madgwick filter freq = sensor frequency ) get me the best result.

Thank you,