Question AirSim Drone Unity

Hi! I'm a new on Unity. I have a task to implement reinforcement learning on AirSim. I want to use ML-Agents. But I don't understand how I can embed a trained model after into a real drone? Perhaps someone has already solved a similar problem. Thanks for any ideas.

 

It's an ONNX model, which can be run with pytorch, tensorflow etc. You "just" need to give the same observations at inference as you are giving during training. Obviously this will be more challenging than it sounds. What are your inputs? Is it rays? Well, you don't have rays typically on a real drone... Are you using visual input? This will take longer to train, but will still look nothing like the real world input from an actual camera in the actual world. I feel your biggest challenge will not be embedding the trained model into the real drone, but trying to get the inputs in the game to somewhat resemble the inputs in the real world.

 

Thank you for your answer! Yes, I have to think about this task. I will experiment. Something should work out in the end )