IIRC the machine sorts like that already, but you can't have a bin for every separate type - you'd need a warehouse of boxes, and the boxes would have to be of several sizes. So I think the pictures just illustrate how he's chosen to cluster for now, as an example to people who try to decide how they'd actually want their purchase to be grouped.

Correct. A typical bin contains many different part ids but it would be trivial to make other divisions.

Neat!

So each item is actually identified right down to the size, colour, type, etc, but then literally bucketed into a group of similar items?

Yes, because otherwise I'd have to make very many passes through the machine to sort a particular batch.

I was thinking that you had trained the identifier just on the bucketized classes instead- fewer classes comprising all the similar parts. Maybe higher identification rate that way, or maybe not. Identifying right down to the specific colour and part means a heck of a lot of classes to train though.

This kind of stuff fascinates me. I've worked on software for package sorting machines in Amazon warehouses. Very similar idea to this: identify, remove from conveyor at right place/time. Only the machines are millions of dollars, run at very high speeds, and use barcode scanning for identification.

I can only dream of the kind of setup that you could make with a large budget. My only advantage is that I can put a lot of time into this project.

One approach that seems obvious is to re-sort based on orders. If someone wants 30 different piece types, but all in blue, that'll be a pain. But if someone wants any subcategory of what's already identified, that could be approached by pulling the existing box for a new sorting run.