Realistic simulation of multiple UAVs collaborative inspection of large outdoor structure

Abstract

Algorithm testing for unmanned aircraft requires significant resource requirements. Simulators are solutions to the resource-demanding methods for evaluating algorithms before they are deployed on a real unmanned aerial vehicle (UAV). They help users to reduce time and increase repeatability during simulations. Despite the abundance of simulators on the market, they are fragmented in meeting the user requirements to perform algorithm testing. This project aims to embed virtual perception sensors in a monocular visual-inertial system (VINS) UAV to inspect virtual outdoor structures during a real flight. The proposed simulator should satisfy the following requirements: (1) a realistic environment containing reliable sensors output and the ability to construct new scenarios for testing, (2) fast and efficient computational processing, and (3) achieve accurate state estimation during simulation. To develop the simulator, a rendering engine is implemented to allow users to create realistic environments for their simulation and a state estimator will be implemented to optimize the visual inertia odometry process for state estimation. To test the performance of the simulator, a motion tracking system was implemented to track the monocular visual-inertial UAV ground truth to compare with the visual inertia odometry generated. The primary contributions of this study, in comparison to previous simulators, are (1) establishing sensor data connection; (2) simulating visual perception sensors to physical UAV; and (3) successful performing inspection tasks with physical UAV.