Hey HN! I'm someone who both flies and programs drones. I've often struggled to make my drones perform perfectly, especially when the environment changes like a gust of wind.
To solve this, I’ve trained ML Models to replace the Control Loops currently in charge of controlling most drones. These models perform better than Control Loops, require no tuning, and retain this performance in many more situations than Control Loops.
Requiring no tuning also means that the same model can be used for multiple different drones/configurations, so it works out of the box.
A high level model is also trained to perform obstacle avoidance and navigational pathfinding.
Check out the demo: https://youtu.be/fWgD7tjZugY
I'd love to hear any feedback or questions you might have! Thanks :)