I love my robot vacuum but I dare never run it when I’m not at home for fear or smearing poop all over my floor on accident. Sadly I didn’t shell out enough money for a bot with superior object detection.
I am curious though if I can model a dog turd out of something and have the robot recognize it properly....
Now you could probably attach a raspberry pi with a camera and a coral tpu to a cheaper robot, spend a week trying to get the TPU to work with its dependency hell (or use one of those cameras with object detection built in), and then probably not have much more control other than to pause the robot just in time. That might end up costing more than just to upgrade to a robot with poop detection that can automatically circumnavigate them.
I like to experiment with home automation things, and recently I added a feature to my system where I could push notifications with images to my phone (pretty much everything I do is entirely self-hosted btw). Then a Valetudo update came out - Valetudo is a FOSS replacement for the cloud services usually required for supported robot vacuums. You root your robot, block it from phoning home, install Valetudo, and now you have a robot that won't spy on you unless you specifically tell it to.
So the update introduced a feature where photos with obstacles recognized by the robot are available via the UI/API. So I put something together where I'll get the pic sent as a notification immediately.
In effect, it's like I've subscribed to the most boring instagram feed of the world. Just random low-quality pictures of items in my apartment. The low quality and weird perspective actually makes it look intentional or filtered, as if a mouse is being forced to make a collage for photography class but doesn't feel inspired at all.