This is the most amazing DIY robotics content I've seen on YT in a long time, and that's saying a lot as it there's definitely no lack of it. A unique approach that even people without formal education in control theory can appreciate.
I think your accuracy is better than that. Going tip to tip like that the test indicator ball is round so if the x or y is off a bit it will show up as more error in z. The reason being the tip of the pencil is very sharp as well so it's a point load on ball. 2 balls are hard to measure in the z like u have set up. To remove that extra error in measuring, use a small plate at the tip of the pen so the ball is against the plate. I better your accuracy jumps up 30-50%.
I love the encoders, it's absolutely lovely! I did a similar thing on a setup where the system would not turn more than ~350 degrees, by printing a radial gradient that swept around the center of the paper disk, so only one sensor was necessary to read the position. Since it did not go full circle, though, the problem of crossing over the boundary was not solved, and I like your method much better! I do wonder, though, if the magnetic-type encoders might be useful as well? (as I understand these involve attaching a small magnet to the center of the motor's shaft, and a hall effect sensor or magnetometer is placed in line with the magnet to detect the angle at any point)
There are two encoders on each servo. The two IR sensors (counts as one encoder) is used on the output shaft of the motor while a magnetic encoder is used on the output shaft of the gearbox.
@@TheMaidenOnes Ah, so they're already familiar with the magnetic ones, then! I've seen some incredibly tiny ones before and thought it might be a useful suggestion :)
I did try exactly what you suggested first, the problem I ran into was that the magnetic field of the motor interfered with the as5048a I used. But the code still allow using a magnetic encoder instead of the optical one. So if you or anyone else find a good way to mount it on the motor it is easy to test ;)
Is it possible to use sensorless approach for motor speed? e.g. using current ripple counting? Or perhaps better to use FOC BLDC? So that position can be known even at low speeds?
You need stepper motors... micro stepper motors. I mean how many times are you going to mod a motor. Its a great project but not very repeatable. Awesome work...subscribed.
In theory you could, if you know all the forces and the exact backlash you could do an open-loop control of the backlash. The problem is that that approach will never be as robust as if you actually measure the position.
@@adamBackstrom4 Oh, I made a bad assumption that forces are negligible compared to the stiffness of gears/motor, but it seems i was wrong. Nice, thanks for explaining :)
naive question: is it possible that the calipers themselves are acting like springs and absorbing a bit of the backlash? And it's a good thing that you use an off-center paper circle because all of mine end up that way..
Not as far as I can tell. The spring only act in the z direction but, as @Lucas Is Busy pointed out, this setup isn't just measuring in the z direction.
This is the most amazing DIY robotics content I've seen on YT in a long time, and that's saying a lot as it there's definitely no lack of it. A unique approach that even people without formal education in control theory can appreciate.
Amazing!
Damn!
Impossible!
Adam Bäcklash pushing everyone's buttons with stunning accuracy.
You are inspiring me to try make impossible possible
This is amazing. I just found your channel, and you're on another level.
I think your accuracy is better than that. Going tip to tip like that the test indicator ball is round so if the x or y is off a bit it will show up as more error in z. The reason being the tip of the pencil is very sharp as well so it's a point load on ball. 2 balls are hard to measure in the z like u have set up. To remove that extra error in measuring, use a small plate at the tip of the pen so the ball is against the plate. I better your accuracy jumps up 30-50%.
excellent - great design and amazing accuracy. Well worth all the effort and work you have done - brilliant.
After discovering SimpleFOC a few years ago, this is the next thing to fascinate me in actual control systems!
Pure Magic. I'm totally gonna read into this.
Thank you so much. Truly Amazing work.
I will be turning your design into a liquid handling robot for use with a micropipette
Sounds like an interesting project! Looking forward to some feedback on the build instructions ;)
I love the encoders, it's absolutely lovely! I did a similar thing on a setup where the system would not turn more than ~350 degrees, by printing a radial gradient that swept around the center of the paper disk, so only one sensor was necessary to read the position. Since it did not go full circle, though, the problem of crossing over the boundary was not solved, and I like your method much better! I do wonder, though, if the magnetic-type encoders might be useful as well? (as I understand these involve attaching a small magnet to the center of the motor's shaft, and a hall effect sensor or magnetometer is placed in line with the magnet to detect the angle at any point)
There are two encoders on each servo. The two IR sensors (counts as one encoder) is used on the output shaft of the motor while a magnetic encoder is used on the output shaft of the gearbox.
@@TheMaidenOnes Ah, so they're already familiar with the magnetic ones, then! I've seen some incredibly tiny ones before and thought it might be a useful suggestion :)
I did try exactly what you suggested first, the problem I ran into was that the magnetic field of the motor interfered with the as5048a I used. But the code still allow using a magnetic encoder instead of the optical one. So if you or anyone else find a good way to mount it on the motor it is easy to test ;)
Very cool, I need to try this on my projects!
Wow! That is so sick!!!
"Build your own."
^^^ THIS ^^^
Seriously impressive!
It's awesome, keep going.
Very interesting! 😊 Have you used Arduino or RPi Pico? 🤔
So... how accurate would this be on a "regular" system without that much backlash? 😳
People who start off an always use the best of everything sometimes never learn how to make something from nothing.
Awesome work.
Is it possible to use sensorless approach for motor speed? e.g. using current ripple counting? Or perhaps better to use FOC BLDC? So that position can be known even at low speeds?
After looking at the "mituitung" I started to question which one is more accurate, the robot or the dial
It's the dial.
just wow. amazing work...
how many grams can it hold in extended position? thanks
You need stepper motors... micro stepper motors.
I mean how many times are you going to mod a motor. Its a great project but not very repeatable. Awesome work...subscribed.
Incredible!
Really great!!!
Backlash is constant size for each servo? Couldnt you just use 1 encoder and include backlash in the code for when the robot changes direction?
In theory you could, if you know all the forces and the exact backlash you could do an open-loop control of the backlash. The problem is that that approach will never be as robust as if you actually measure the position.
@@adamBackstrom4 Oh, I made a bad assumption that forces are negligible compared to the stiffness of gears/motor, but it seems i was wrong. Nice, thanks for explaining :)
naive question: is it possible that the calipers themselves are acting like springs and absorbing a bit of the backlash?
And it's a good thing that you use an off-center paper circle because all of mine end up that way..
Not as far as I can tell. The spring only act in the z direction but, as @Lucas Is Busy pointed out, this setup isn't just measuring in the z direction.
Neat!
The music is torture.
I liked it. Gave the video a futuristic vibe.
Also like it
I wouldn't put it that strongly but it would clearly be better without it. Any futuristic tone gets dated pretty quickly too 🙂