[krayvonk]

Heres a paper about this cool idea.
https://ieeexplore.ieee.org/document/6942770

Proprioception is a kind of "invisible perception of your own body"   We have it many ways, not just a cerrebellum, we actually can remember it as well from last seeing it, and guessing where it is.

If you dont have encoders in your robot,  you can just use plain old vanilla dc motors and dont worry about half the circuit coming back to the computer.
Its easy to add encoders,  but if you had an encoderless software, but still had the encoders for backup,  then youd gain fault tolerance with it.

I think that if you induce an environment model off a track history,  this may be one of the ways you can develop this "invisible sight" to the body that it cant see, but can just feel the bumps on the head, shaking the camera.
It involves having something else tho,  perception, even perhaps an off view from one of its muscles (like a head maybe.) to work it out,  so you cant have it for nothing,  one thing that I know is that context is a bitch,  without enough context (or clue) not even miracle worker can give you a straight answer without reading your mind.

So off I go to work with my next demo,  not sure if its going to get me anywhere, but im confident its around the corner, every corner.   

[HS]

All you need is some way, any way of collecting info about the body. Theres a documentary about a guy who got nerve dammage from polio I think. And he can only walk/be coordinated when he can watch his body.

But this means that if the power goes out at night he cant walk until morning. So some detection methods are better than others.

[krayvonk]

In the form of a sensor, or a memory of a sensor, indeed.