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

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