Distributed Deep Reinforcement Learning for Drone Swarm Control

We applied Deep-Q Reinforcement Learning, with the use of CNNs, to manage a drone swarm scenario in a bi-dimensional environment. Given a square map with some targets, the goal is to make the drones able to learn to cooperate between them, trying to track and follow the most valuable targets. We compared a distributed and a centralized approach and verified how the first can outperform the latter in a real-world scenario with limited training.