Construction of a virtual reality platform for UAV deep learning

Abstract

In this paper, based on Unreal Engine 4 (UE4) Unmanned Aerial Vehicle (UAV) simulator (AirSim), through AirLib, DroneServe, DroneShell and other modules in AirSim, using window10, 64-bit operating system, Intel Xeon E5-2620 v4 eight cores Dual CPU and Nvidia GeForce GTX 1080 Ti (11263MB / Nvidia) dual graphics card, a virtual reality platform is constructed for UAV deep learning. The platform displays mountains, lakes, and trees, and can indicate four seasons change. Through this platform, images and data for artificial intelligence research can be obtained for UAV deep learning, such as UAV autonomous obstacle avoidance research or autonomous flight research. Because of the considerable advances in graphics hardware, computing power and algorithms, the platform is realistic and can accurately show subtle things (such as dazzling sunlight, mirror reflection of lakes, and elastic deformation of trees). This virtual reality platform performs in real-time, and there is no delay in image and data. Thus, it can meet design indicators of the virtual reality platform for UAV deep learning.