Back when I was playing with DIY Drones, there were a bunch of projects using optical mouse sensors to do visual position hold at low altitudes. There was a fairly well known way to swap out the lens, then you'd get a 16x16 or 32x32 pixel "image" stream, which was good enough to do feature detection and at the same time made that feature detection way less computationally expensive than hi resolution video feeds.

I strongly suspect anything a robot vac does with cameras could easily be done with super low resolution sensors. Even if you needed to put 2 of them in stereo to get depth perception that maybe you could compute from HD video.