Complementary Filter - Sensor Fusion #2 - Phil's Lab #34

[@tobbe3344]

Thank you a lot. I found an article on the internet about the Kalman Filter last year. That artice helped me a lot in realizing and understanding the Filter for an attitude estimation system. Now, I realize that this very article was yours. So I want to thank you. Keep up the good work! I am excited for the video about the Kalman Filter.

[@iamnarval]

I have been planning to get my head around Karman filters for a while, so I am very exited for the next part! Very nice explanation in this video. The state observer interpretation was especially enlightening

[@nutkickermotioncontrol8238]

Thank you! Very nice explanations. Quite fast-paced but just about managable for my brain.

[@PhilsLab]

Thank you - I’m glad the pace was just about okay!

[@daphoosa]

Good introduction of a core concept needed for attitude estimation. For less powerful microcontrollers there are functionally equivalent filters with significantly less computational cost (no trig functions) , but as a stepping stone to the EKF, this is perfect.

[@socialogic9777]

Dear Phill. One day I will start watching your videos and watch all of them

[@chrislamb4723]

Another outstanding demonstration showing thoery in action! Thank you!

[@sarbog1]

Cool..... will need some time to wrap my head around this!!! THANK YOU!

[@thundrwaffle]

Love your videos on filters and other signal processing topics! Im currently doing a Masters in signal processing so this really is awesome

[@wizardOfRobots]

Curious why yaw rate is not mentioned, only roll and pitch rates?

[@nabeelsherazi8860]

This is such a cool series I can’t wait for more, thank you Phil!

[@yasinbedirhansimsek2883]

Perfect video in every way, Looking forward for the next part

[@madmonkey7183]

This is awesome. It's a really excellent explanation of the system works. Thanks. I also really appreciate making the source for everything open. Now to apply it to rockets 😀

[@mustafaefecetin]

Excellent work, very comprehensive

[@gsvachan]

Quick Question: How are the raw accelerometer readings and raw gyroscope readings being filtered in real time? what is the computation behind the values stored in "lpfAcc" and "lpfGyr"? And is the Low pass Filtering of the raw values necessary? PS: Loved your video, the only video that didnt talk about only theory.

[@u1trathunder]

Fantastic explanation and perfectly timed. Just started a robotics project that could really use this info. Keep up the great work!

[@islammohamed3954]

Thanks alot for the valuable content. Please keep the series going.

[@Andres-is8zz]

Excited for the next part! Thank you!!

[@dmitrynuzhdin]

Cool! I will be waiting for the Kalman filter video! Do you plan to cover Madgwick and Mahony filters as well? Very good job!

[@chadkrause6574]

You explain things so simply. One thing I don’t understand though - it seems like the complimentary filter is basically just 5%*accelerometer data + 95%*gyro data. If that’s the case, why don’t you still get drift from the gyro?

[@kevinvermeer9011]

Gyro drift is still present at the input, but you're integrating the gyro data with the combined data which includes the accelerometer compensation. As long as the gyro drift effect is weak or small enough, the accelerometer contribution overrides it.

[@mecitpamuk5623]

@@kevinvermeer9011 I guess because of the highpass filter to the gyro and lowpas to the acc eliminates drift

[@iwbnwif]

Thank you for the clear, concise explanation. One thing that I missed was whether this type of sensor integration also copes with linear acceleration - for example along the aircraft's longitudinal axis - or is it only for rotational accelerations? The reason for asking is that I think a longitudinal linear acceleration might be falsely sensed by the pitch gyro as a pitching movement.

[@martinmckee5333]

This will not handle linear acceleration correctly. The gyroscopes will be unaffected, but the orientation estimate from the accelerometers will include the influence of the linear acceleration and, as such, will generally be incorrect. This can be corrected if the acceleration vector is known - simply subtract the linear acceleration from the accelerometer readings. Alternatively, the gains of the complement filter can change when acceleration is present so that only the gyroscopes are used for tracking orientation. Of course, that causes gyro drift issues again, but if linear acceleration is rare, it can be workable.

[@b21-soaring]

LOL I just recognised the Cambridge crsid in the github repo. Calling roll Φ (or was that acceleration) seems anachronistic in an era where everything is software (i.e. the Φ looks great on a blackboard in the Baker Building, but "roll" looks better in C). Although you could double down with vars called Φ_gyr_rad.Nice job with the video series though.

[@dymastro788]

Thank you for this, quick question: Any update on the paid course?

[@kevinvermeer9011]

Looking forward to the next one - it seemed obvious that you'd want to dynamically update alpha: when you're sitting still, the accelerometer is more trustworthy, the gyro is just drifting, and you'd want a large alpha, when you're pitched over in a turn or otherwise changing quickly the accelerometer is less useful and you'd want a small alpha.

[@martinmckee5333]

Yes. It is true that gain scheduling is beneficial in the case of dynamic motion that includes linear accelerations.

Thank you for sharing.

[@cosmicazur]

Can't wait for KF implementation and link to serial oscilloscope is not in the description mate

[@gankankg]

Great video again 👍👍

[@wizardOfRobots]

Great video. Thanks!

[@krishnawa_]

You will make a PCB course right?

[@safayetkhan2754]

Thanks a lot! When you will upload part 3? Thank you again!

[@PhilsLab]

Thanks for watching, Safayet! Part 3 will come in the next 2-3 weeks.

[@pointlessanonymous7960]

i hope i'll see you post part 3 asap cuz i have an exam about this tomorrow...help

[@lukaswalczak93]

Could you make a video on different integration methods such as trapecoidal rule etc?

[@robstoddard9521]

Why did you not use quaternions?

[@MEan0207]

It's really useful. 😀😀😀😀😀😀

[@bilalzubair4248]

Sir I learn very much from your explainations. Can you please make a video on how to interface a Tamagawa or any other resolver with stm32 for motor control applications. It would be very helpful.

[@RicoElectrico]

I wonder if that could work for double integration (e.g. a rail car moving on a straight track - fusion of accelerometer and GPS). Obviously just for fun, not as a practical application.

[@PrasannaRoutray97]

Is Quaternion Kalman in the works?

[@nhlakaniphombatha5769]

👍👍👍❤ thanks a lot

[@alihancoban]

I don't understand the filtring part I think you should use high pass filter for gyro datas .. am I wrong?

[@PhilsLab]

Imagine you are rotating at a constant rate (i.e. DC) - what is a high-pass filter going to give you in that case?

[@mecitpamuk5623]

@@PhilsLab But in this case after some over the time, gyro drify will be huge effect because we didnt apply the high-pass filter? Am I wrong?

[@akanguven114]

L.e.g.e.n.d God bless!

[@YoursTruelyMe2]

I know these videos take a lot of time but if you ever wanted to take time off and write a book, I would back it.

[@obregr]

interesting

[@PhilsLab]

Thanks!