Distributed Circle Formation Control for Quadrotors Based on Multi-agent Deep Reinforcement Learning

Multi-rotor unmanned aerial vehicles (UAVs) have been increasingly used in both civil and military applications in recent years, and it is difficult for a single multi-rotor UAV to accomplish a specific task due to the load and range limitations. In this paper, we take the most commonly used quadrotor among multi-rotor UAVs as the research object, and study the formation control problems such as formation generation, formation transformation and collision avoidance. We design a distributed circle formation algorithm based on deep reinforcement learning, which is combined with an artificial potential based law to ensure collision avoidance. In order to simulate the real environments, a multi-quadrotor simulator is built based on ROS and gazebo. The algorithm is trained and tested in the simulator, whose results show the effectiveness of the proposed algorithm.