I want to make a litter cleaning robot

I want to create a robot that can pick up litter and this requires detecting litter nearby and differentiating between litter (metallic objects, plastics, glass and paper), natural objects (sticks, leaves, trees and bushes) and walls. Does anyone know how to do that? Thanks in advance!

To clarify, I expect multiple sensors, not just one. Also, this category was the only sensor category I could find, if there are any other sensor categories, let me know please!

I doubt that anyone can do that without very good cameras, computer vision, artificial intelligence and a substantial computer. Even then, there are probably humans in the loop to separate the last bits. If you figure out how to do it cheaply and reliably with an Arduino, patent the process before you post it. You could make millions.

How many years do you have to complete the project ?

However many is needed, this is just a thing that I want for my own nonprofit (clutter crushers) and a robot would ease up on human labor

There is not yet, a 'one sensor to rule them all' that could detect that mix of stuff.

Vision recognition is perhaps one way to do it, see here;

But forget running that on an Arduino.

I think the only way to differentiate materials don't make a lot of sense here.

  • Tap the object and use AI to tell what the material is based on the sound it makes. There's a model of Porsche that uses this method to determine if you are driving in the wet.
  • Use ultrasonic sensing to somehow figure out how much of the ping actually gets picked up, determining the object's sound absorption and therefore what it's made of.
  • Take a photo every few milliseconds and then reverse image search it to figure out what it is using the keywords that come up.

I never expected ONE sensor, I should probably add a note on the post to clear that up

1 Like

This is really really really hard.

The sanitation department in NYC claims that they are making great progress in processing single stream recycling and are able now in their multi-million dollar plant to separate many different materials. It would be instructive to check into how they do it.

On the other hand, there are still plants just processing glass that employ humans to identify and manually remove any detritus on the conveyer.

Doing this with an Arduino is a really big ask, you're likely going to need cloud based machine learning algorithms to make progress.

Right, take a photo and use image recognition to look for lettering. If it has any lettering, of any kind, on it it is probably man made. Depending on the environment that could be very effective. Say in a forest pretty much everything that is man made doesn't belong. Whether it belongs where it is or it is litter is another problem.

With a camera system like that it is not an Arduino project any longer.

1 Like

OP would never be able to run this on Arduino. Maybe a Pi 4 2GB?

And then yes, the issue you mentioned is another problem.