A novel leader replacement-based unmanned aerial vehicle flocking scheme

The flocking motion is a fundamental and crucial operation in multi-UAV systems, encompassing navigation and obstacle avoidance. However, the unknown and random environment poses a great challenge to traditional Unmanned Aerial Vehicle (UAV) flocking and navigation control methods. This paper uses Reinforcement Learning (RL) techniques to achieve navigation and obstacle avoidance for a swarm of UAVs in unknown environments. The RL algorithm employed in this study is MAPPO, which has been integrated with an attention mechanism to form the ATT-MAPPO algorithm. This incorporation of the attention mechanism enables the agent to effectively filter out irrelevant information and focus its attention solely on crucial features relevant to the task at hand, thereby significantly enhancing decision-making accuracy and efficiency. The leader replacement strategy effectively balances the energy distribution within the UAV flocking, thereby significantly extending the overall flight range of the UAV. Finally, we demonstrate the scalability and adaptability of ATT-MAPPO in a simulation experiment.