The Octo-Bouncer: Advanced Bouncing Patterns 

Thanks for mentioning that part. Wasn't easy to pull that off.

@@ElectronDust A great example of how something so simple can rely on a lot of work but it pays off for sure!

@@ElectronDust I've seen tutorials for Blender on how to track the live footage and then program the render camera to match. What did you use?

@@JohnDlugosz It's actually running in real time. I am using Unity (the game engine) to render the moving 3D model. I also programmed all the image processing and data visualization in Unity. So when the machine is doing it's thing the 3D model moves as well. I just recorded both my screen and the actual machine and then edited the footage into one video.

Plot twist. It was ALL a simulation 🙈🙊🙉

Found the platform bouncer connoisseur :)

I was thinking the same, it never loses control. That's so cool. Reminded me of an entrainment video with metronomes. Does rythm has something to do here?

And that’s how spacex is able to land rockets....

@@ps-gq5km science is badass

@@ps-gq5km by bouncing them a few hundred times first?

Oh my god. That's a great idea. Thank you!

@@ElectronDust or one of those wooden tongue drums with a rubber ball. nice work!

I was kinda hoping for that

ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-6CrVADr44TU.html 🤣

I listen to rhythm trance music .

thanks!

@@ElectronDust Did you use c or matlab to code. Prob both but I'm clueless.

I like your idea. And I've thought about it. It should be possible.

That would be so cool

@@ElectronDust I am like to see that you reply to the comments on your video.

How much does all equipments cost ?

Good tennis players do that all the time. No joke you can huck a fastball at them and they can catch it with their racquet like a glove.

I agree. By the way nice channel

Well put

thank you!

Lmao

I'd like to see that too.

@@ElectronDust about two balls on the same bot ?

It would be especially cool if each one had its own camera and could only see the ball when it was near, not when it was at the other end. It can know where the other one is, but not see it. A variant would be to see and track the companion platform.

Or just have the one bouncer, but attach it to linear rails so it can translate.

Many great ideas in this thread! I am currently working on two-ball bouncing, the difficulty at the moment is to do the up-down movements fast enough in order to be on time when alternately hitting one of the two balls.

why though

@@davicarvalho6699 For the lulz

thank you!

I agree.

well, you will still need to do some math to get the correct xy position of the ball. As you know, things closer to any camera-like device appear bigger and move "more" than things further away. What I mean by that is this: If you move an object 10cm away from the lens 2cm to the right and measure the difference in pixels in the image data you will get a far bigger value compared to when you move an object 100cm away by 2cm

Nice idea! I like it.

The hardest part was getting the Camera data stream running in Unity. I had to write a custom C++ plugin which uses OpenCV libraries (just the parts to get the camera data stream though.) And all that just to get 120 FPS camera image data and be able to set key settings of the camera device (exposure, shutter speed, gain, ect.) That was quite the hurdle.

In responce to the Analytical tilt control: Yes, I wrote all the control code in this project.

I see, thanks for the information. I use OpenCV in just python to do image processing on digital microscopes (1 MP or less) and rarely get more than 20 FPS. I've tolerated it because it's not critical to get such a high frame rate, but wow! 120 FPS. Maybe I should switch to Unity if I want to get serious about it.

I had to write the circle detection algorithm myself to get fast and accurate results. There was a circle detection algorithm in OpenCV (using an algorithm described as HT-21 in some papers) which also worked. But it was slow. It uses the Hough Transform, which is nice, I love the Hough Transform. It was just overkill for the specific image data I got with my camera. Overkill in the sense that the image data was way to clean. It didn't need anything fancy like the HT-21 algorithm to detect the circle in the image data. Maybe you are able to improve performance on your image processing as well by re-evaluating the algorithm in place. Maybe you are able to delete some redundant step. Maybe you've already done this. I imagine you spend a lot of time tweaking your code. Keep also in mind that my image data is only 640*480 (0.3 MP.)

@@ElectronDust Interesting, I see what you mean. So your system would have to adapt if you wanted to do it in a noisier environment. What is the bare limit for FPS where this can work? Ever tried using something like a rubber ball instead? I would really love to see a few machines like this "juggle". Seriously, this is one of the coolest things I have ever seen. I wonder how applications could be served by a machine like this, but this is what you chose to do with it. Really, really, way to go.

Probably yes, because the motors actuate based off the ball's instantaneous position. (It's closed loop)

I love this indepth hint. Thank you very much. I didn't think about the uneven wear on the gears at all.

I haven't. Great idea. Thanks for the tip!

That would be neat. Love the idea.

Hey, I had no idea what voice coil motors even are. Thanks for the tip. Looks promising!

Sure! The big trade off with commercial ones is cost, as far as I think, but maybe if you tweet at h2wtech or something they'll sponsor you. I know if I saw this video I would.

thanks for the hint!

I was wondering how hard it would be to get a plate to bounce a ball. The target really was to satisfy my curiosity.

Things people do during quarantine: ● Isaac Newton (1666) - optics, laws of motion, gravity and calculus ● William Shakespeare (first lockdown, 1592) - write poetry _Venus and Adonis_ and _The Rape of Lucrece_ (second lockdown, 1606) - write _Macbeth_ and several other of his best plays ● Titian (1576) - painted _Pieta_ before succumbing to the plague himself ● Edvard Munch (1919) - painted _Self-Portrait After The Spanish Flu_ ● Giovanni Boccaccio (1351) - wrote _The Decameron_ ● Electron Dust (2020) - built and coded ball-bouncing robot paddle

Yes. That's my next goal.

@@ElectronDust Make it 3 balls, start juggling, and break the internet

Second ball will retry first ball...

Good thinking. I was puzzled by this as well. I was double checking that the machine really moved the way it's suppose to move. And it is. At one point I suspected that some of the energy on hit might get stored as angular momentum and this would cause the ball to abruptly bounce in an unexpected direction on the next bounce. But this wasn't the case either. Currently, I think that small irregularities in the surfaces of both the ball and the plate as well as dirt are the cause for some of the unexpected bounce behaviors.

It's using analytical control. The idea is that because we know the balls current position and velocity, and we are furthermore able to get a good estimate were the ball will hit the plate, we should be able to calculate the ideal tilt by just applying basic Newtonian mechanics.

One is fairly accurate. We're talking +/- 2mm height accuracy. Though the depth perception gets worse towards the edge of the screen (not horrible though, and this could be corrected in software). It gets worse because of the camera's lens distorting the ball's shape more towards the outside. I think 2 would be overkill for this specific application.

@@ElectronDust would it be useful to remap the image to correct the lens distortion?

@brutesquefilms well yes, but I don't think I'd get 120FPS anymore with that kind of image data remapping. It might be better to just correct the height data whenever the ball gets towards the edge. It feels like there's a clever solution somewhere along this line of thought.

First I just wanted to get my steppers to drive very fast and smooth. That's important. If we aren't able to get fast and smooth moves we won't get anywhere later on. This is one of the main pillars this project stands on. Next, we need to work out the IK of this specific system. This takes some time and a lot of checking, but it is also very important. No accurate moves without 100% valid IK equations. Another important thing was to get 120 FPS image data running in Unity. This was harder than I thought. I had to write a C++ plugin which utilizes OpenCVs camera interfacing code to get this running. But this is another main pillars this project stands on. Than there's the circle detection algorithm. You guessed right; Another main pillar, jep. And the microcontroller-PC communication protocol; Main pillar. So after all that stuff was in place, I finally could start to get some control algorithms going. I used PID first, because it is very easy to implement. After that I implemented gradient descent to get better velocity data (the position data is a bit noisy, that's why we're using gradient descent to get a good approximation of the slope of the position curve, i.e. the velocity.) After that I looked at a lot of data and thought about why the PID isn't performing as good as I thought and how this could be improved. I noticed that with all the data about the ball, I should be able to predict were the ball will be landing. So I added hitPositionDetection. Using this and all the other data and basic physics I implemented Analytical Tilt Control.

@@ElectronDust thank you for the explanation!

Maybe a table tennis practicing machine

Hey, thanks! It's Unity

no, it's all just code. No AI involved.

@@ElectronDust There is some moments with more punch at start, the ball goes really high The game is to stop the ball in the middle of the glass if I've understood... Good challenge this works very well