GSoC Reinforcement auto rotation


I had an idea to work on in GSoC for this year.

I would like to make an actor-critic reinforcement model to train a neural network to act as a controller on the copter to do the auto rotation safely, it would be extended to do other missions but that’s what came to my mind first.

Once the neural net could do pretty well in the simulation, it would be copied as a function to the copter controller, i think this would save alot of computational time and power onboard

This cincept can be generalized for a lot of things in the project…if there’s any suggestions or comments please let me know