One of the many projects I have pop in my head repeatably is an automated litter picker. In my home neighborhood, there’s normal city litter combined with the spare sharp. In my work neighborhood, there’s everything. I started by thinking about how to dispose of difficult items like needles and dog poop bags, but I think I need to start by figuring out how to identify the litter.
We have a lot of categories, some of which we don’t want to touch on quite yet. I think what’s be best is start by building a rover with a few sensors. The first iteration should be able to follow me around the block, taking continuous stills, ideally normal, IR, and whatever else. These stills should ideally be pointed out to the sides, and not straight ahead. I can then use these images to start training some image identification. While that’s going on I’ll start looking how to build the more in-depth mechatronics. Once I have the mechatronics done, I can then reach out to the city about how to pick up and dispose of the more difficult of the dead skin cells of the city. I don’t want to be on the hook for poop bags and needles. I see enough of them that I know there’s a need for safe and clean disposal, but not enough that I’m ready to die on the slopes of Mount Filth.
The major things that I think I’ll be identifying are size, color, proximity and liveliness. If it shows up on IR it might be living with bacteria. If it is, is biodegrading and the problem is solved. I don’t want to build anything large enough to pickup a pallet, let alone a stool. There should be no weeding, only litter. Glass will be a weird one, do I want to puncture plastic, etcetera.
The start though has to be identification. It’s like the classic Rumsfeld trope: “There’s the knows, the known unknowns, and the unknown unknowns. It’s the last ones that get you.” I probably butchered it, but I thought he was an ass outside of a couple good sound bites, so meh. If a thing isn’t positively identified, it should be flagged for the next round. It shouldn’t be touched unless it needs to be.
Then it’ll be motion. I still really like the Godzilla concept. Either that or a trash picking up scorpion. A scorpion would be appropriate (Scorpio here) and easy to do system in terms of balance, but it will be a lot of motors which reduces the deployability. I don’t know how will it would climb stairs. I could think of using the tail to help push it, but could be a definite weak point in the design. Godzilla would be fewer bigger motors, maybe, but would definitely suck on doing the balance being fixed with the tail.
I’m leaning scorpion the more I think. Stairs are going to be a challenge either way, and a scorpion will be more applicable to other future needs.
Admittedly, I could go sideways on the design, but somedays straight and easy sounds pretty good. I do like the idea of the visual-behavioral gestalt.