"yet" ? Deep learning is no panacea - it's not clear why this would even be on a top 5 list of approaches to try on a putative ML part of the process. There seems to be a lot of confusion on this, generally.
I'd think once you have built sensors that can digitize the whatever properties of the plastic that are relevant to sorting them, actually recognizing the different types of plastics would be rather simple.
Sure, it works fine if you're selective about the plastics you use, especially if you avoid using objects that were made using a mixture of different plastics.
Honest question, what's likely to happen if you mix plastics? I understand how if melting points are different you might get odd results, and I'm guessing maybe very different materials won't bond at all leading to cracks in the final product. But if you doing something like they show in these videos, heating it to <200 degrees and just pressing the result into sheets or blocks that can be cut with hand tools, or doing very simple injection molding, can't you get away with a lot that might not be acceptable in an automated, professional setting?
Note that I'm thinking entirely in terms of what I might do myself in my workshop, not what'd scale to a city-wide or country-wide recycling system.
Well, I know that some thermoplastics can be mixed with good results. I've a roll of polycarbonate-ABS in my printer right now. Normally polycarbonate prints at 310C, and this blend prints at 295C.
Ive seen other blends as well, with intermediate temps between the 2 initial plastics. Now, I only have experience with PLA, ABS, Nylon, PVA, Polycarb-ABS, and HIPS. A lot of this is still unexplored as it took industrial quantities to do anything.
WHat these machines will allow is experimentation on small batches and how plastics work and don't work. And that's awesome.
Most likely they just use cheap labor for sorting