Відео

Well, you can check out the video about the TVC model rocket. The attitude estimation/correction system was built with quaternions and formulas I have covered. And it's probably can be considered "robotics" :) Same goes for any drone / UAV or for example describing orientation of the robotic arm end effector.

lol

If you are talking about the board, the next version will be available. The one in the video has issues. And I will make a video about it when it will be ready.

Thank you! I will try to make more content. I have a lot of material for videos, but unfortunately making videos takes a lot of time (sometimes I feel like it is harder than the project itself lol)

Go for it! Starting could be hard. I probably spent a year thinking about the rover, before I started actually doing something. I was scared of the amount of efforts required to finish it (and I still not very close :) ). But when I started I was making progress way faster than I though. And it is also fine to take breaks if you feel burned out. So I would say that if you have an idea and you are passionate about it, just go for it. The faster you start, the faster you will get to something you will be proud of. Good luck!

Yeah! That's sounds cool! How big is the robot? I am working on the next version of the servo. The version in the video has some issues. I will share design files when it will be ready.

This encoder returns absolute value :) Yes, the zero position is quite random, because you cannot really align the magnet's poles orientation when attaching it. But if you don't mess with the magnet, zero point stays the same, so you can just offset it in the code.

Well, I am not a professional, so I am trying to figure it out myself :)

Yes! The host controller can read position and other stats from the servo

@@Positive_Altitude Nice! How do you set the I2C address on the servo side?

@@yeroca for now it is just hard-coded, but actually STM32 can write to it's own flash during execution. So I think that I will reserve a place on the flash, and make a special I2C call that sets a new slave address. So you take a servo, make that call, reboot the servo and now it has a new address. But it's a plan for now. I never touched that feature before, but if I read the reference manual correctly, that should work.

@@Positive_Altitude Sounds like a workable plan.

1) англоговорящая аудитория намного больше 2) так как контент инженерный, даже среди русскоговорящей аудитории подавляющее большинство понимает английский 3) для тех кто не понимает ни один из этих языков и пользуется автоматическими субтитрами (таких около 30%) перевод с английского работает значительно лучше

Yeah, I will use CAN bust in the next version. I think Ethernet is a an overkill here. And CAN actually an industry standard to connect things in cars. This rover also has way more torque than it really needs, but it will be slow :) But I am fine with that. I already made a board that does a battery management and has MPPT solar panel charger, but it is designed for 10V.

'tmp' stands for 'temporary'. This formula is an approximation and it makes the norm to slightly increase with each step. I mentioned that rotation quaternions should be normalized, but actually, you can use non-normalized quaternions with the same math to transform coordinates. But in this case, you will not only rotate the space, but also you will scale it by the factor equal to the quaternion's norm. That's why if we let the quaternion's norm increase, the object gets bigger. To explain why exactly this happens I will give you a very similar example: Imagine that we are trying to draw a circle, but we can draw only short straight lines. And we use an algorithm: 1) Draw a line from the current point to the center (a radius) 2) Draw a short line perpendicular to this radius 3) The end of this short line is the next point of the circle 4) Go to 1) If we do this with very short lines this kinda works, but actually, every new point will be slightly further away from the center, and the radius will increase with each step. And if we keep doing it we will draw a spiral instead of a circle. If our steps are small it will spiral out slowly. With bigger steps, the problem gets worse. But we can fix that. After each step, we can make a little step toward the center that will make the current radius equal to the desired radius. That is exactly what normalization does. This example is VERY close to what happens with quaternions and this formula. In the case of quaternions, it's just a trajectory (not really a circle) in 4-dimensional space which I personally struggle to imagine. No, this is unrelated to renormalization. This is just us fixing an error caused by using approximation.

@@Positive_Altitude Thank you, I think I'm getting it. The normalization sort of re-orients or self-orients the origin point of the gyroscope, the center of all the angular velocities right? That's why when you turned normalization off, it sort of dilated in-and-out/breathed like a growing spiral?

@@Positive_Altitude Do you need to renormalize all quaternions or only gyroscopic ones with angular velocity? Does angular velocity add more requirements to make it work?

@@erawanpencil I would not say that it "re-orients" it just scales it. Norm is literally a length and all proper rotation quaternions should have the length = 1. We just fix the error caused by this formula that makes quaternion slightly longer than it is supposed to be. Usually, when we just combine rotations there is no need to normalize quaternions. Because if we have two normalized quaternions, their product will be a normalized quaternion too. But also there is no harm in normalization. If you normalize a quaternion that is already normalized it will not change it. The angular velocity formula is just kinda weird. Mathematically it is obtained with the assumption that time step dt is "infinitely small". But there is no such thing as an infinitely small number when we do 3D graphics or iterative simulations. So in real life, this formula produces a small error both in direction (orientation) and length (norm). To deal with that we need 1) use as small dt as we can, to minimize the error in orientation 2) normalize quaternion to always keep norm = 1 as it is supposed to be

I am working on the video 👍

Thank you! Not shared on GH yet, because there are issues and no documentation at all. Maybe I will do that during the next few weeks. But anyway I am working on a new, improved version, that will be shared from the very beginning. I hope I will get there in 1-2 months.

Yeah, that's right. I did not want to overcomplicate things and said that is just a PWM, which is almost true. But I got your point, thanks!