I am trying to build a machine where I can dump assorted Lego pieces into a big hopper and then the machine will take one piece at a time and decide into which bin it should be placed and put it there then repeat.
The machine will have to identify thousands of different types of pieces but sort them into just 10-20 categories. How this will work is that we will sort into a few main categories and then additional pieces will be sorted out only when needed. A simple example would be if we are running the machine to sort into the categories of green, red, blue, white, and other and then we realize that we need 120 yellow 2"x1" pieces for a project we can have just the yellow 2"x1" pieces sorted into a 6th bin in addition to the machine continuing to sort into the other 5 categories.
There are indeed many Lego databases available online but I am not sure whether these will be usable; maybe as the very base of the machine learning.
Here are the main questions that we will need detailed answers to:
How do we get the machine to take one single piece at a time when some pieces are 20x as large as others? (this may be partially solved by presorting different sizes with various grades of sifters)
How do we get the one isolated piece to be identified by the computer based on size/shape/color and hopefully type of plastic if possible? (AI Computer Learning and various scanners/cameras)