UAV Autonomous Control Using Multi-Agent Reinforcement Learning in a Cooperative Control UAV Network

As unmanned aerial vehicle (UAV) technology advances, the use of UAVs as temporary networks in disaster areas has been considered.It is possible to provide a network for the disaster area by forming a relay network between UAVs in the air disaster areas where power and communication networks have been disrupted.However, because each disaster area's environment differs, deploying UAVs that form the relay network needs to be designed for each location.The authors have proposed cooperative control of UAVs using multi-agent reinforcement learning and have been investigating autonomous control according to the environment.This report discusses the arrangement of UAVs according to the relative positions of the landmarks.Multi-Agent Proximal Policy Optimization (MAPPO) is the algorithm for multi-agent reinforcement learning.Computer simulation results clarify the autonomous movement of UAVs.