Reinforcement Learning from Gaming to Auto-Landing

Unmanned Aerial Vehicles (UAVs) are revolutionizing numerous industries, yet the complexities of executing missions in dynamic aerial environments present significant challenges. This study explores the application of Deep Reinforcement Learning (DRL) agents to address these challenges, specifically focusing on the autonomous landing of drones on maritime vessels. While conventional systems allow for automated landings, they often lack adaptability to specific locations and orientations. In contrast, DRL agents excel at learning optimal control strategies tailored to intricate tasks and diverse conditions. By mastering these strategies, DRL agents can significantly enhance the safety and efficiency of autonomous landings, paving the way for advanced high-level API integration in landing commands. This research aims to demonstrate that leveraging DRL can transform UAV operations, making them more reliable and effective in challenging environments. Ultimately, the findings have the potential to not only improve maritime operations but also extend to various other applications where precision and adaptability are paramount. This innovative approach positions DRL as a key player in the evolution of autonomous UAV technology, driving advancements in operational capabilities and safety standards. To support this research, we created a simulated environment using Unreal Engine 4.27, UAV provided by AirSim plugin and RL training facilitated by Gymnasium and Stable Baselines3.