This week I stepped back from the app and asked a bigger question: how does the brain actually talk to a body. NEURA has lived on my own hardware as a companion you talk to, but the desk robot was always the plan, and this week I started building the road between the two.
The first piece was a hardware abstraction layer. The idea is simple even if the work was not. NEURA should be able to say "look left" or "roll forward" without knowing what motor or board is underneath. Swap a part later and the brain learns nothing new. It sends the same command and the layer below handles the rest.
On top of that I designed a small internal protocol so the three players, the brain, the face, and the drive base, can speak to each other. Nothing fancy, and that was on purpose. I wanted something tight and predictable that I could reason about at three in the morning when something will not move and I need to know why.
Then came firmware. I wrote the code for the microcontrollers that run the eyes and the motors, the chips that turn a message into an actual blink or a turn of a wheel. It is the least glamorous part of the week and probably the most important.
What I kept telling myself is that the contract matters more than the parts. Getting the shape of these messages right now, before there is a full robot to argue with, saves me a mountain of pain later.
It still mostly lives on the bench. But for the first time NEURA has a nervous system, even if the body is not all here yet. That felt like a real line to cross.