We’ve been working on an interesting project for a few months now (it feels like an eternity) so I thought of writing an update.
Some of my friends and I have a little “girlfriend” Plexiglas We used to do some consulting work in the past, but now mostly use only as an umbrella for projects.
After spending like a year on and off working on RaaS (Reality as a service, it was a web page that could be sent to OpenAI fitness jobs and run on real hardware) We decided to move on to small-scale projects on purpose.
The first was a new friend https://www.philipzucker.com/markov-slackbot/.
Next was a marble mirror. The idea is to draw a picture using colored marbles somehow. It was inspired by the mechanical mirrors of Daniel Rosin http://www.smoothware.com/danny/ Very cool.
We were crazy to think we could do it in a month (meet once a week). Like it’s actually funny.
But we have reached a milestone!
meaningful event! pic.twitter.com/fWjtwTHqRN
– Philip Zucker (@SandMouth) October 28, 2021
This is a laser cut marble mirror. At the bottom of the marbles there is a “connection 4” mechanism that is activated by servo. “Pacman” moves the bnalls in the column. Rear 2 steppers and servo. The stepper controls a carriage with an RGB sensor and servo. He paints a picture of his choice (4×4 at the moment) in silver and blackened steel bearings. We have a python code that rejects bullets until it gets one that it can use. Github code is here https://github.com/perciplex/marble-mirror.
it was a long time ago.
Now how the hell do we rate?
The post on Ben’s blog about the project http://blog.benwiener.com/electronics/projects/programming/2021/10/28/marble-mirror.html
- End of May 2021 – Beginning
- June – 3D printed prototypes. A lot of propensity for ideas and discussions goes nowhere.
- July – Purchase a laser cutter. Mess with it a lot. Helix design
- August – Beyond the Deklon method. Another technology similar to a 3D printer. Carriage with swivel screw + steps. Pit bull fever bypasses Plexiglas. Is that ironic?
- September – Super busy month. Ben’s wedding is imminent. I’m moving. Did not work much
- October – Parade of victories!
We started to see what was out there in terms of marble control.
You can choose to use marbles or steel bearings. You can have such bags outside of Amazon. Sling ammunition seems like an interesting option, but it is not particularly funded. Our current approach is to use a steel blue liquid (super blue) on chrome bearings. With this we can get black matte balls and silver balls. Hot Tip: Apply Multiple Layers, Clean First With Denture Alcohol. We were a little confused at first why the coating looked so bad.
A big question was how “multiple” our control over marbles is going to be. Some degrees of active freedom (DC and servo motors) to use, sensors. This seems to be a common solution that the more passive you are going to be, the more precision you need in your manufacturing and tinkering.
Choosing a large-scale design for the screen
Some initial designs were tested on the 3D printer. This revealed problems in the process. A fundamental problem with shaping the “gut” is that you need shallow slopes, otherwise there are massive gaps between each row. This is bad for the look and quality of the image. The 3D printer created separate ridges on these low slopes. Maybe these were objects of choosing the movement of the printer head? It is not clear.
Some convincing early prototypes.
I really liked this design. The idea was that we could use stacked three-dimensional printed rails to create a coil. If the front parts are made of clear acrylic, in principle there is no need for spaces between the balls. An out-of-place traffic problem arises. The bullets did not move reliably in this direction and could get stuck. In the end it was decided that we were trying to be too smart. I still think there’s something here. Design documents: https:
Crazy version. What if the orbit is a giant spiral? Then rotating the entire follow-up moves the marbles inward. Never really expressed himself
Release Vertical Rows + Connect 4.
That’s right now what we’re going with. This is very natural
We bought basics 40W laser cutter July 2 as a leap of faith. It felt like we were very limited by our ability to create a rapid prototype. In principle we could send a cut every week, but it did not work and money was always a function forcing not to do so. It appeared and worked! In the end I am very happy with it. We followed the instructions online to try to add air support. It was Bondogel who spent weeks on the faulty behavior of laser cutters. We returned the stock parts and it seems to be working just fine for our needs. We also added an ammeter on the front. The time Ben and I drilled through steel and bent and cut tiny pieces of sharp Chinese steel was truly awful. In the end, we have already turned on the laser at about 18mA recommended at a reading of 55%. Oh well.
We can cut 1/8 plywood at a speed of 8 mm per second at 55% power. Sometimes it does not really work and needs help with an x-acto knife.
We use K40 Whisperer as our control software. You can create an SVG or DXF vector drawing and import it. Whisper color K40 encodes the lines to decide whether to engrave or cut. We basically use the onshape program drawing to export dxf after removing the extra Crud it draws by default.
Techniques for building laser cutters:
Originally we had a homemade CNC router machine and 3D printers. In the end this was a big problem. The CNC machine was simply unreliable, slow, terrible to be near and difficult to use. 3D printers are very flexible but very slow and not very dimensionally accurate at all without fine tuning.
To identify marble in the carriage we use Ephroth RGB sensor
It’s a bit strict and not entirely clear that we should continue to use it. We painted the inside of the carriage blue with a paint band. For now, it seems to have taken some examples of an empty ball, a black ball and a silver ball and then using the classification closest to the neighbor from scikit-learn works fine.
A smart trick was to use two back panels so that the screws would not interfere with the balls and rails.
We have a cheap raspberry pie motor driver board. It behaved strangely. not worth it. Buy a real Edprey.
It was nice to get real points of contention. Also handle screws are curled for easy disassembly