Deep Reinforcement Learning Based on Curriculum Learning for Drone Swarm Area Defense

In a drone swarm (also known as a UAV swarm), coordination is an important research topic in swarm control. In this paper, the problem of a drone swarm defending a target area is taken as the background of research on the cooperative control of the drone swarm, and deep reinforcement learning is used to form the control policy. We construct a detailed environment model about drone swarm area defense, including intrusion policy of non-learning drones, capture process modeling, and Markov Decision Process (MDP) modeling. In addition, we propose a new reinforcement learning algorithm with curriculum learning, in which different numbers of agents performing policy communication correspond to various courses. Finally, we compare the proposed algorithm with other algorithms experimentally, and the results show that our algorithm has better policy performance.