stuffmadehere

@scottbez1 How thick are they? You could maybe do something similar to coin dispensers - a spring loaded stack of cards where only one can slide sideways

I'm building a machine that will pick-and-place plastic cards, but they keep sticking together when lifted. How do I avoid this? My attempted workarounds - shaking back and forth and scraping against some foam - aren't reliable enough. See picks 3 & 4 in video...

I committed to writing A LOT of cards, so I made a robot to do it for me. Then I wanted to know if they were good so I sent them to a handwriting expert: youtu.be/cQO2XTP7QDw

I need your help! If you know anyone who is a forensic handwriting analysis expert who would be willing look at some interesting writing & be interviewed please DM me or email at stuffmadehere@gmail.com

@spreen_co Still plan to release it! Just gotta find time to clean it up and document.

@stuffmadehere still waiting for the data set of the puzzle pieces as a coding challenge 🤐

If you liked my painting robot you should check this one out. It took him over a year to get it working and has some really nice results that play to the strengths of the painting process. youtu.be/09LlZhlcJTU

@oscarmayer it has come to my attention that you have a fleet of weinermobiles including a weinermini and weinerdrone. I see a big gap in your lineup just waiting to be made. I think what the world needs is a walking weinerdog. Please make #projectdachshund a reality.

@joe_puzzo That said, more accessible robot arms would advance other industries that would use them to reduce labor costs. For example having a robot arm tend a CNC machine to load parts in and out allows a single person to make parts 24/7 without hiring employees.

@joe_puzzo I don’t think they’re comparable. 3D printers allow you to make almost any part with little effort. This is insanely useful for most makers. Robot arms are neat but I don’t think many makers have much use for them like 3D printing. So, no, I don’t think it’s a big issue.

@stuffmadehere 3d printers are now accessible for many people, opening many new doors. However, robotic arms are still out of arms reach for the average tinkerer. Your typical 6axis cobot that can lift 2-6kg costs on average 28,000 USD. Do you see this as an issue?

I sat down with the my old boss @MaxLobovsky and discussed my time working at Formlabs, tools and other miscellanea. Check it out here if you’re interested: youtu.be/ztW5ywbh7FU

I spent a really long time making a robot solve a 4000 piece all white puzzle so that I had an excuse to talk about software algorithms. Check it out here: youtu.be/WsPHBD5NsS0

@zbzzn I did it many times with smaller puzzles. First a 9 piece, the. A 45 and they didn’t have the issues that the big puzzle had.

@stuffmadehere But why did you scan the puzzle 4 times instead of using, say, 500 pieces puzzle until the algorithm works? I almost cried when you had to do it the second time.

Does anyone watch the @stuffmadehere videos? I keep getting the feeling he always tries to do stuff the hardest way possible just to test out the limits of his equipment and not really solve a problem.

Just got done doing an interview with the Formlabs CEO (@MaxLobovsky) from my garage. Mostly talking about my time at Formlabs and digital manufacturing. You can check it out at the user event Formlabs is having this week (it’s free) summit.formlabs.com

@PennJenks @FilipoGiovanni Yeah I’m doing ransac, but the issue is mostly garbage input (the edges of pieces are often “torn” / not sharp after die cutting). Even if you get perfect scans, some pieces are cut very similarly. Like they fit perfectly if you put them together, but it’s wrong.

@stuffmadehere @FilipoGiovanni I thought that might be a problem -- but just in case, are you using a robust model (curve) fitting method like RANSAC or L1 objective?

I thought a jigsaw puzzle robot would be a nice easy project. Nope. Although to be fair I chose to make it as hard as possible… If you want to see what I came up with, I posted a video about it here: youtu.be/Gu_1S77XkiM

@FilipoGiovanni @PennJenks I tried that, but many of the edge pieces can fit in multiple locations. If you get it wrong it cascades into much wrongness in the interior.

@stuffmadehere @PennJenks Do you know what percent of the best fit edges are wrong? If you do the outer edge first and then fill in the inside in a circle, you’ll always need to match on two edges when you place a piece. And you’d immediately get to check the third edge when you place the next piece.

@PennJenks @FilipoGiovanni The challenge is that due to error in image acquisition / piece cutting defects, the edge that’s most similar may not be the correct edge. Thus a simple local / greedy algorithm doesn’t work (at least for me). You want all the possible matches then do a global best fit

@stuffmadehere @FilipoGiovanni Is that really necessary? A KD tree will work great for finding nearest neighbors with this scale problem. Approximate nearest neighbors (which I think is what you're proposing) has the added challenge that it's sometimes wrong

@FilipoGiovanni @PennJenks There’s a special kind of hash called a locality sensitive hash which unlike a typical hash tries to make similar quantities collide. They’re great for this kind of thing.

@PennJenks @stuffmadehere You’d have to quantize the curve. There might be some collisions, so it wouldn’t be perfectly constant time, but you should be able to engineer a pretty reliable representation.

@mcdanlj Those puzzles give me strong NP hard vibes (if you’re not using the picture)

@slightedlime I actually bought and tried to use a vibratory bowl feeder. It had a number of tricky problems that made it not worth it. As for sorting, its necessary to properly deal with cumulative error. If you build two disconnected islands, they generally won’t connect properly.

@stuffmadehere Is it possible when it scans the pieces to assign them a number value (i.e. 1,2,3,etc.) and then when it solves it puts the numbers in the correct order so that it doesn’t have to sort the pieces again before putting it together? Also, auto-piece-feeder.