In ground reconnaissance missions, high-altitude UAVs typically encounter enemy air defense zones or cloud interference. Meanwhile, high-altitude UAVs are generally flexible and lightweight with large wingspan, so the influence of wind is hard to be ignored. For the obstacle avoidance flight task of high-altitude UAVs, a controller based on deep reinforcement learning is proposed, which can output angle and thrust commands according to the current state, enabling the aircraft can avoid obstacles and fly to the target area. For the design of the controller, a wind field model is constructed from the wind field data obtained from measurements, a three-degree-of-freedom aircraft model is built based on a certain type of high-altitude UAV, a training environment is set up by combining the wind and aircraft model, and a reward function is designed based on expertise. The simulation results demonstrate that the trained controller can avoid random no-fly zones and fly to its destination in real-time, and can adapt to the changes of the wind field.
Autonomous Obstacle Avoidance for High-Altitude Unmanned Aerial Vehicles Based on Deep Reinforcement Learning
Springer Aerospace Techn.
International Conference on Advanced Unmanned Aerial Systems ; 2024 ; Nanchang, China September 20, 2024 - September 22, 2024
2025-03-18
13 pages
Article/Chapter (Book)
Electronic Resource
English
MOVING OBSTACLE AVOIDANCE FOR UNMANNED AERIAL VEHICLES
British Library Conference Proceedings | 2013
|MONOCULAR VISION-BASED OBSTACLE DETECTION/AVOIDANCE FOR UNMANNED AERIAL VEHICLES
British Library Conference Proceedings | 2016
|Obstacle Avoidance for Aerial Vehicles in Autonomous Navigation
British Library Conference Proceedings | 2021
|Obstacle Avoidance for Aerial Vehicles in Autonomous Navigation
TIBKAT | 2021
|