Numi v2: AI Coin Sorting

AlbumNerd

After playing around with my AI coin grading project for half a year, I've started to sour on the idea. I think Technical Grading will eventually be accurate enough to be useful, but I don't think the market will accept it. PCGS, NGC, and CAC are just too dominant. People want and trust the judgment of these companies. I just can't see AI grading taking off in a big manner.

But...after testing AI over a thousand times, I noticed that it's extremely accurate in identifying coins 99% of the time. Specifically identifying the series, year, and mintmark. [For US and World Coins, not medals or tokens. Nor varieties/errors]

I've been working towards a robotics system to identify and sort coins using ROS2, a Raspberry Pi, a mini camera for the AI to take and analyze images, and Legos bricks. Consider it Numi v2.

The goal is to build a coin-sorting robot that uses AI to intelligently sort coins. Imagine a reverse Coinstar that spits back out valuable coins.

My friends working at coin shops lament about all the bulk coins coming in that they don't have time to sort. Such as wheat cents and silver world coins. Which is a bummer because they're leaving a bunch of money on the table.

I'm starting with wheat cents, but the goal is to work towards sorting other US coins, silver world coins, and eventually identifying varieties & errors.

Attached are images of the initial designs I built using Lego Studio Designer. I'll be building the physical prototype over the next few weeks.

What would you want out of a coin-sorting machine? How could this machine help you or your shop?

PeakRarities

If I had a shop, I would absolutely love a coin sorting machine. I don't have a shop, so I can't give you any actual feedback as to what it would need to do 🙃.

erwindoc

If you had the volume to justify the purchase of a machine like that, Im sure it would be worth it.

ColoradoCoinGuy

Ideas to add. Using Lincoln cents as an example. Load in the dates/mm you want to sort for. Example 1909-S VDB, 1909-S, 1914-D, 1922. Have the machine sort through the thousands of cents feed into it, spitting out the desired dates. The rest fall into the reject bin and are also counted. Even though you say AI grading may not ever happen, you still need some way to determine if a coin is say XF or better, as some early Lincolns, say 1912-D, are worth keeping if XF or higher. These also would need to be sorted out.
Also as @erwindoc eluded to, you need to be able to sell a machine like this pretty cheap, as most LCS might not have the volume to make this machine worth it.

Not trying to hijack your thread, I think AI grading will eventually come to pass, especially for grading circulated coins.

Keep up the good work, Justin.

AlbumNerd

Work on the Numi v2.0 prototype is going strong. I'm still waiting on a few remaining Lego pieces, but it's coming together. It was fun bringing this to my local coin club tonight to get feedback.

DelawareDoons

For what it's worth, something similar has been done. There's a reason wheat cent bags were running up near $250+ a few years back... Dude built a machine that would sort out the better dates. Drove the whole market on bulk wheats up. IIRC he went bankrupt because he couldn't sell his leftovers fast enough since everybody knew they were trash and he stopped finding enough good stuff to make it worth the premiums.

ColoradoCoinGuy

@AlbumNerd idea is to sell these to coin shops so they can use it so sort through all the Lincoln Wheaties they buy over the counter. You would be supprised at how many Whitman albums have the wrong cents in the wrong holes. Employees don't have the time to check out each book for keys that are in the wrong hole. Same goes for "junk" silver that is coming in now in bags as silver prices rise. A lot of these haven't been searched for keys and doing a $300 bag of Mercs by hand is not economical. If he can make one of these machine to sell for a low cost, he might just have something.

Thanks for bringing it in to the club.

AlbumNerd

@DelawareDoons said:
For what it's worth, something similar has been done. There's a reason wheat cent bags were running up near $250+ a few years back... Dude built a machine that would sort out the better dates. Drove the whole market on bulk wheats up. IIRC he went bankrupt because he couldn't sell his leftovers fast enough since everybody knew they were trash and he stopped finding enough good stuff to make it worth the premiums.

Whoa! Do you know the name of the guy or his product? I'd love to learn more

AlbumNerd

Update! Numi v2.0 alpha prototype - 07/04/2024

After a few weeks of tinkering and coding, I've built the initial prototype and flow for Numi v2.0. The prototype can capture images on both sides of a coin, send an API request to GPT-4o for analysis, and display results identifying the coin's type, country of origin, and year.

Video of Numi 2.0 in action
Screenshots

Tech Specs

• Computer: Raspberry Pi 5
• Camera: 2 x Arducam IMX519
• Software Language: Python Code
• Physical System: Lego Build Hat peripheral and Lego technic parts
• AI Model: GPT-4o & OpenAI API

Below are some of the problems encountered, lessons learned, and what I plan for the next iteration.

Problems

• Maintaining consistent lighting is a major issue. Lighting is the #1 determinant for accurate results. I am struggling to find a powerful light source that can connect directly to the Raspberry Pi that lets me control both light intensity and color.
• The total time for a coin to go through from placement to results is around 30 seconds. This is too long. It takes around 12 seconds for the API call to complete with the rest for the coin to manually move through the system.
• An API call to the GPT-4o model takes around 1 cent per coin analyzed. Costs can add up fast if you're sorting thousands of coins.
• The AI is struggling to identify mint marks. I suspect this is due to inconsistent lighting as I know that GPT-4o can identify mint marks just fine when I upload close-up photos from my phone.
• Finding the right camera has proven extremely difficult. Most Raspberry Pi cameras have major limitations. It's surprisingly difficult to find a camera that can capture clear images within 50cm of an object, has autofocus, AND zoom functionality. My IMX519 cameras requires the coin to be at least 5cm away, it does have autofocus, but no zoom so I have to manually adjust the cameras.

Lessons Learned

• Initially I thought writing the software was going to be the biggest challenge, but instead, it was the mechanical engineering and building the machine itself
• Linux' permissions setup for folders is the bane of my existence.
• Building with Lego pieces never gets old
• I did not need to use ROS2. Python scripting works just fine.
• The first design of a robotics system does not have to be perfect. Better designs come from making mistakes and seeing the system come to life.

Plans for the next iteration

• Mechanism to sort coins based on certain parameters
• Goal of reducing run time to 25 seconds per coin

Long Term Goals

• System to pull coins out of a pile and place them onto the machine
• Capture and store images in a database
• System to place coins into 2x2's and staple it shut
• System to print key info on 2x2's such as coin type, year, mint mark, silver content, etc
• Explore open-source AI models to reduce costs

Based on lessons learned, I've already designed what the second design iteration should look like. I've compartmentalized the coins and cameras from the sorting system. This way taking photos and analyzing coins will not be a bottleneck for sorting the coins. The Lego parts are being mailed and I aim to build the next iteration in a few weeks.

I'll be at the Summer FUN show this Thursday - Saturday. I'm hoping to chat with a lot of dealers & collectors to better understand how they currently tackle identifying and sorting bulk coins. The goal now isn't to sell Numi or convince others that it's valuable. But rather to understand the needs of collectors and dealers and see how AI can make their lives easier.

If you'd like to share your thoughts at the show, let me know which table you are at and I'll drop by! Even better if you have some Dansco albums for sale

Manifest_Destiny

tcollects

this is amazing, I'm interested in your progress

abbyme24

Can't really add much from a discussion or technical standpoint here but just want to say I am super impressed with what you are doing!!

yosclimber

It looks like a fun project to play around with, especially since you like Legos.

I believe some of the long term goals are much more difficult than you can do with Legos.

• selecting one coin from bin and placing on conveyor belt for photos & IDs
• stapling a coin into a 2x2; this would include rotating the coin before doing this

This means there would be lots of human handling involved.
And in this process, the human might as well skip the machine and identify the coin faster and more accurately than your machine can.

And @DelawareDoons already explained why the economics for pulling better dates from bulk wheat cents did not work,
even though a guy was able to build an accurate machine to do it.

yosclimber

@DelawareDoons said:
For what it's worth, something similar has been done. There's a reason wheat cent bags were running up near $250+ a few years back... Dude built a machine that would sort out the better dates. Drove the whole market on bulk wheats up. IIRC he went bankrupt because he couldn't sell his leftovers fast enough since everybody knew they were trash and he stopped finding enough good stuff to make it worth the premiums.

Was Tim Rathjen the guy who did this?
https://www.youtube.com/watch?v=k7okDtRRCcY
https://coinauctionshelp.com/forum/viewtopic.php?t=23904

hummingbird_coins

@yosclimber said:

@DelawareDoons said:
For what it's worth, something similar has been done. There's a reason wheat cent bags were running up near $250+ a few years back... Dude built a machine that would sort out the better dates. Drove the whole market on bulk wheats up. IIRC he went bankrupt because he couldn't sell his leftovers fast enough since everybody knew they were trash and he stopped finding enough good stuff to make it worth the premiums.

Was Tim Rathjen the guy who did this?
https://www.youtube.com/watch?v=k7okDtRRCcY
https://coinauctionshelp.com/forum/viewtopic.php?t=23904

That intro is so cringe.

AlbumNerd

@yosclimber said:
It looks like a fun project to play around with, especially since you like Legos.

I believe some of the long term goals are much more difficult than you can do with Legos.

• selecting one coin from bin and placing on conveyor belt for photos & IDs
• stapling a coin into a 2x2; this would include rotating the coin before doing this

This means there would be lots of human handling involved.
And in this process, the human might as well skip the machine and identify the coin faster and more accurately than your machine can.

And @DelawareDoons already explained why the economics for pulling better dates from bulk wheat cents did not work,
even though a guy was able to build an accurate machine to do it.

Legos are fun and versatile, but I'm hitting the practical limits that you mention. The way I'm thinking about this is to use Lego technic parts to build a proof of concept.

If I can pull out off, even in a suboptimal way, I can justify to myself getting a 3D printer and printing custom parts.

My definition of success is if I can run my prototype and generate a personal profit sorting out coins

gumby1234

I think sorting by hand is much quicker and it doesn't cost you 1 cent per coin. Sometimes I sort coins while watching a movie.

AlbumNerd

@gumby1234 said:
I think sorting by hand is much quicker and it doesn't cost you 1 cent per coin. Sometimes I sort coins while watching a movie.

One of the things I'm looking to talking about with dealers at Summer FUN is to see how the economics of automated sorting would work.

One way to reduce cost is to lease the machine rather than sell. That way shops don't have to have a large up front cost.

I also wonder if the time saved would make it worth it. Assuming minimum wage is $15 an hour, that my machine can sort 400 coins an hour, as well as costing 1 cent per coin analyzed [so $4 an hour].

What kinds of coins sorted out would pull out at least $4?

Are dealers even paying for employees to sort coins?

If so, what coins are those employees sorting?

Even if Numi can sort 10,000 coins a day, do coin shops even have that many coins to sort through?

AlbumNerd

I showed off my latest build of Numi at yesterday's Denver Coin Club meeting

Latest Changes [As of 7/26/2024]

• Optimized cameras for better focus

• Fully built out the second design iteration after getting all needed Lego parts

• Added LEDs for lighting

• Updated code to handle the entire identification and sorting flow [Capture images, send images and prompt to AI model, move coins to correct position for sorting]

• Migrated AI models from GPT-4o to Gemini Flash 1.5

Current results

• Lighting from LEDs are too weak, yielding inaccurate results

• Cameras work

• Motors for moving the coins beneath the cameras work, but the final motor for sorting is buggy. Needs to be fixed

• Time needed for analysis improved from an average of 30 seconds per coin to 10 seconds

• The Gemini Flash 1.5 AI model is shockingly accurate and blazing fast

Upcoming Changes

• Add high powered lights to brighten up coin photos

• Add and program a coin hopper to fully automate the sorting process

• Goal is to optimize the machine to sort 100 coins in a row with no issues. Get Numi ready in time to show off at the Denver Coin Club Show happening in mid-August. [I got a table!]

My hope is to get Numi into a good enough state that I can challenge attendees at the show if they can sort 100 wheat cents faster than Numi. If they can, they get a coin prize. Should be a fun way to get young numismatists engaged.