Deep Reinforcement Learning Unmanned Aerial Vehicle Autonomous Cruise System with Fusion of Visual Information

• DOI: 10.1145/3660043.3660158
• Publication Date: 2024-05-30T14:18:07Z
• Authors: Rui Liu; Ying Shi; Yufei Xie
• Keywords: Drone; Obstacle avoidance

UAV have a wide range of applications in military, aviation, agriculture, logistics, and other fields. However, when conducting real drone testing, there are high costs, safety risks, environmental impacts.In this case, the virtual simulation scheme automatic cruise based on Airsim Unreal urban environment proposed article can provide an economical reliable solution, which has broad application prospects field research development.This utilizes open-source cross platform simulator provided by to simulate scene information. By inputting parameters actual data into system, it flight control process, effectively test verify performance algorithm UAV.Meanwhile, also easily replicate adjust experimental results, improving testing efficiency reliability.The advantages paper as follows: 1. Visual information coding strategy is integrated deep reinforcement learning algorithm, solves defects traditional processing visual certain extent. 2. A multi-information fusion Reward function distance reward, position-angle-deviation punishment obstacle avoidance items designed introduced, make adapt more complex changeable physical environment.