Machine learning excels at classification, and trash sorting is at heart a classification problem.
Applying ML to non-classification problems is when it is NOT appropriate, but this is a good pick.
Applying ML to non-classification problems is when it is NOT appropriate, but this is a good pick.
Classifying by physical properties is even easier by simply classifying by physical properties. Want to pull all ferrous metals out of a pile of stuff? Just use a magnet. Want to sort tin from plastic? Blower and conveyor bins by flight distance.
Incidentally, those elements not easily covered by this kind of sorting (which has been used for many years) is also not easy to cover by magical do-all machine learning, which will have a hard time sorting plastic-coated paper from paper from plastics, just like the processes above.
ML is useful when you cannot make a clear difference using physical properties only. For example, if you treat different types of plastic garbage, you will probably want to separate bottles from other types of plastics, and you can do that with visual sorting and pattern matching.
> which will have a hard time sorting plastic-coated paper from paper from plastics, just like the processes above.
Certainly, but then this is when you still need some manual work at some point of the chain. But physical filtering + ML could reduce the human part of the work to exceptions.
(0) It isn't hard.
(1) If the trash bins are aligned in a predictable manner, then you can even do it without thinking.