Kalman Filter for Beginners, Part 3- Attitude Estimation, Gyro, Accelerometer, Velocity MATLAB Demo

[@תפארתנבו]

Thanks! After watching some videos and be confused, your tutorials made Kalman filters clear finally..

[@ProfessorRoss]

Glad it helped!

[@eigd]

I just watched the three parts in one go, working on a final test in our "probabilistic robotics" course, where we are to fly a drone through a maze. Let me first say thank you for a very pedagogical, well structured and easy to digest manner (having had digital signal processing helped building the intuition from your perspective using the averaging filter approach, this time around, it is from a probabilistic perspective using bayes rule). I agree with your view that, unless you are looking to write a phd on kalman filters, you don't need to look into how it is deduced. We just let linear systems (or linearized ones) and gaussian assumptions do their thing, hehe. What I found though, is the application of different sensors, and your process states to be estimated, and how to produce A, and H matrices. I am currently using the Tello drone, looking at the API and scratching my head on how to eventually produce an estimate of the state of the drone. There is a barometer, TOF sensor, IMU, and even velocity estimates from a closed, proprietary visual positioning sensor that I don't have a clue about how accurate it is. Not to mention the prediction step, before doing control, and then planning and trajectory following... What really breaks my mind, is the multiple level of uncertainty. I can for certain say I am uncertain about my understanding about what is going on. Anyways, thanks for a very enlightning course :) Cheers

[@michaelmina2311]

I've been trying to learn Kalman Filters for a while with no luck.. After watching these videos, things are coming together. Thank you very much for these excellent videos.

[@MasudRana-cf9fn]

Really great presentation! You explained all steps one by one with theoretical and practical phages. Billion of thanks from my side!

[@ProfessorRoss]

You're very welcome!

[@malithjayalath6417]

Thank you very much! It made lot easier after watching your explanations.

[@rahulsb5746]

Hands on the best explanation for Kalman filters of all the explanations that I've seen! It would be great if you could explain or point out resources where we can learn other types of filters and how they are used.

[@ProfessorRoss]

I wish I could, but I'm a complete novice. I knew just enough about the filters I mentioned to teach about them. Maybe after a few years, I'll have more experience with filtering.

[@ElTurbinado]

i know so much more after watching these 3 parts thank you!

[@ProfessorRoss]

Happy to help!

[@maddiesal3270]

Best video series! Thank you

[@alihosseiniroknabadi4828]

Very Well defined. Thank you professor. Wish you health from Iran.

[@ProfessorRoss]

Thank you, and thank you for watching. Good health to you as well.

[@pragotipranbora9018]

Thank you for the excellent presentation. I have a question regarding the Kalman filter MATLAB example for the case of without using accelerometer data (time 27:00). Here IgnoreGravity = 1 and the psi, theta and phi are initialised to 0. Therefore , in this case the measurement 'z' that is passed to the function EulerKalman(A,z) always corresponds to psi = theta = phi = 0. However, for the estimation step in the Kalman filter algorithm we need to provide a newly measured dataset z at each time step. How is this handled for this example?

[@vacoff2717]

great tutorial, enjoyed both 3 parts

[@ProfessorRoss]

Awesome, thank you!

[@Ivanovichx]

Thanks for the great video-series. I just have a question regarding the MATLAB Kalman filter using both data from the gyro and the accelerometer. If I understand it correctly, z is our measurement. Ignoring gravity, our measurement is omega_1, omega_2 and omega_3 given by the gyro. That is converted to quaternions to be able to write x_{k+1} = A * x_k where x is the state with the 4 quaternions. The kalman filter uses H as the matrix that maps states to measurements. With H = identity(4) means we're measuring directly the quaternions (I'm assuming because we can translate omega vector to quaternions. But when we incorporate the data from the accelerometer, I don't see how this fits into the Kalman filter. If I'm understanding the previous video correctly, we have two new measurements, meaning H should be a 6x6 matrix. Instead, the code seems to simply use roll and pitch obtained from the accelerometers as initial guesses rather than measurements. Could you clarify this? Thanks.

[@virgenalosveinte5915]

These videos were amazing, thank you so much. Very comprehensible

[@ProfessorRoss]

You're very welcome! Thanks for watching.

[@pk_star7862]

Best explainnation of kalman filter and its application

[@ProfessorRoss]

Thank you!

[@nicholasrahaim2407]

Would this method work on orbit? My thinking is that the assumption that inertial gravity is aligned with the Z axis would not apply for an orbiting body so you'd have to use some sort of RIC frame kind of like with gravity gradient stabilization analysis. I guess it would be more math and you'd have to use your position estimate to update the gravity vector, but then you'd potentially be able to estimate the yaw angle as well? would love to know if my thinking is way off here, thanks, love the channel!

[@hoopschwitz]

Great presentation, thanks a lot.

[@ProfessorRoss]

Glad you liked it!

[@salvodippolito6013]

Thanks a lot for this three part series! It has helped me a lot in understanding the main concepts of the Kalman filter. What I don't really understand is what kind of operation is occurring within the filter so that it's able to get an estimate of velocity from only distance measurements without computing a derivative? Do you have any insights on this mechanism?

[@ProfessorRoss]

No, I don't know the theory of the Kalman filter. Maybe eventually I will. One does need to have the correct model matrix -- that's all I know. As many of these comments have pointed out, most who teach the Kalman filter focus so much on theory that several students don't learn the practical importance, so I took the opposite approach of just focusing on using it.

[@vimalrajayyappan2023]

Great Lecture sir. Just a small question in the first section,estimating velocity through position, may be I'm new, X = Ax x = [pos,vel] A = [1 dt 0 1] This is the prediction model A for kalman filter. We are using for velocity estimate with just providing the measurement of position alone. How Kalman filter estimates the velocity, because no prediction is there for velocity also measurement has no velocity input considered as H neglects the same. How its estimating velocity from position? Curious!

[@rahultheytv5347]

Thank you so much for sharing, practical information

[@ProfessorRoss]

You’re welcome

[@isabelsoares4315]

It was the best explanation about the Kalman filter I've ever seen, could you tell me which books you used to put together the presentation? I need it to set up a project and all the documents I find are very complex because I'm not in the electrical engineering field Thank you for everything @ProfessorRoss

[@ProfessorRoss]

Thank you. My main reference is the book, "Kalman Filter for Beginners: with MATLAB Examples" by Phil Kim (Author), Lynn Huh (Translator), 2010, www.amazon.com/dp/1463648359

[@isabelsoares4315]

@@ProfessorRoss I looked for the book to buy near me, but I only found it in the USA and shipping is very expensive. Doesn't this book have an online version? Is there any chance you can continue the videos covering high pass filters, with Laplace or Fourier transforms? It would be very important to me! =D

[@eiliyamohebi9701]

Hi, Thanks for your great videos. If we want to estimate yaw angle from gyro we are facing a drift, can we use a bias in kalman filter to estimate this time varying drift and compensate it without using a magnetometer? Thanks.

[@ProfessorRoss]

That's a great question. I honestly don't know the answer, as I'm a beginner myself to Kalman filter use. Maybe I'll have a better answer the next time I teach it.

[@MohamedMounirABBES]

hi , i'm interesting for this question , what is your field study please we can work together ?

[@monkeyones7119]

Hello, I admire your explanation after watching your video. I would like to ask you a personal question, is your attraction pot program drawn with matlab? If possible, can you share it? Thank you very much! Support your videos!

[@kaankutlu1414]

Thank you

[@tomthamos6157]

how could we forecast 100 step into future

[@bluebottle7835]

Thanks for these great tutorial. This is the best explanation on Kalman filter I've ever seen. Quick question on the lecture. Line 28 of file TestEulerKalman.m, the code is z = Euler3212EP([ psi theta phi ]'); Which means measurement z is just the Kalman filter estimation. Is this because the measurement is not available? I initially thought [psi, theta, phi] could be calculated by numerical integration of measured angular velocity like deltaT * [w1, w2, w3] = psi, theta, phi . So calculated psi, theta, and phi could be utilized as z. Please let me know. Thanks in advance