Modular Reinforcement Learning for Autonomous UAV Flight Control

Recently, research on unmanned aerial vehicles (UAVs) has increased significantly. UAVs do not require pilots for operation, and must possess autonomous flight capabilities to ensure that they can be controlled without a human pilot the ground. Previous studies have mainly focused rule-based methods, which specialized personnel create rules. Reinforcement learning been applied UAV flight; however, it does include six-degree-of-freedom (6-DOF) environments lacks realistic application, resulting in difficulties performing complex tasks. This study proposes method of efficient by connecting two different maneuvering methods using modular flights. The proposed divides tasks into simpler tasks, learns them individually, then connects order achieve faster transferring information from one module another. Additionally, curriculum concept was applied, difficulty level individual gradually increased, strengthened stability. In conclusion, were used demonstrate effectively perform realistic, 6-DOF environment.