Hi All,
One of my other hobbies is building Lego. I’ve got over 10kg to sort and I’d like to build something to do it for me.
When I’m sorting by hand I use a multi pass approach, starting with size (using one of these LEGO storage: LEGO sort & store head) then break it down further in to piece type.
In line with this I’m thinking of sorting by weight. They are mostly all the same type of plastic, so bigger parts weigh more. If I can get parts to come through one at a time I could then weigh them and then do direct it based on the weight (set a servo or other actuator to position the output based on a range of expected weights. I can also get a database of piece weights, e.g. a 1x1 round tile is 0.11g and a 2x2 brick is 1.35g
This all means I need a load cell that can detect things in the 0.1g to 5g range but has an accuracy of about 0.01g. Being able to go lower and greater accuracy would be great, but that’s probably the minimum of what I need.
If/when I get this working I’m also thinking of something that sorts identical parts by color, swap out the load cell for an RGB sensor.
Why I don’t want to do image/object recognition.
There are lots of projects out there that sort using image recognition, but I’d rather not based on the following reasons because:
- there are so many different parts and the training dataset is huge, also many different parts can look the same if viewed from certain angles.
- you have to get the lighting just right
- the different colors and prints complicate things further
- given the huge number of different parts, I’m not going to be able to have a bin for each part even if I recognize them.
