By popular demand of one Jeremy, these are notes brought back from more than a decade ago. I didn't start the project as such, but I remember most details and I can expand from there.
Some parts are obsolete, others are becoming incredibly relevant, so don't get stuck on a particular technical detail. It's both a researching and shelved project but there are interesting techniques and approaches, worth writing down and discussing.
It's mostly a hardware project but the NN is a software nut I have not cracked yet, and Jeremy may want to help here.
real-time inference for flight control with supervisor for adaptation/learning.
But this also depends, in the first thousand iterations, on having a real-size wind tunnel (lol)
As you might know, one of my long-time interests is fluids simulations but there are countless parameters to manage... Why increase the complexity to crazy extremes when a NN could find the sweet spot by itself ?
https://www.youtube.com/watch?v=4Qj5y0Vtdds has some nice ideas