A New Trajectory in UAV Safety: Leveraging Reinforcement Learning for Distance Maintenance Under Wind Variations

Abstract

Verbatim from Embry-Riddle Scholarly Commons. Not paraphrased, not summarized.

In the field of aviation, safety is a critical cornerstone, and the operation of Unmanned Aerial Vehicle (UAV) systems is deeply connected with this principle. A thorough analysis and rigorous simulation and testing of aircraft systems are essential to avoid severe safety hazards. This paper delves into the safety issue in UAV operations, specifically regarding maintaining minimum safety distances under fluctuating wind conditions. The study introduces a novel solution based on a Deep Deterministic Policy Gradient (DDPG) model, a reinforcement learning method. The DDPG model was trained using a simulated environment created through the Gazebo simulator, with values for wind and gust conditions derived from historical records at the KLAF airport at Purdue University. The model's performance was evaluated regarding maintaining safe distances under these conditions. The results indicate that the DDPG model can accurately predict safety distances with relatively low error rates when predicting under different weather conditions. The findings significantly contribute to UAV safety operations, suggesting the potential future utilization of reinforcement learning methods to study enhancing airspace efficiency and obstruction avoidance in UAVs.

Authors

• Xu, Xiaolin, M.S. Embry-Riddle Aeronautical University
• Sun, Jeffrey Embry-Riddle Aeronautical University

Keywords

• unmanned aerial vehicle
• UAV
• safety distance
• reinforcement learning
• deep deterministic policy gradient
• flight safety
• airspace efficiency
• UAV fleet operation
• Management and Operations
• Multi-Vehicle Systems and Air Traffic Control
• Navigation, Guidance, Control and Dynamics