Analysis on the performance of VIO according to Trajectory Planning of UAV

Abstract

Recently, Unmanned Aerial Vehicles (UAVs) have been widely researched to accomplish tasks such as structural inspection, drone racing, and exploration of an unknown environment. To make the most of UAV and correctly estimate its pose with inexpensive and light-weight sensors such as Inertial Measurement Unit (IMU) and cameras, Visual Inertial Odometry (VIO) has been intensively researched. There were a few analyses on the performance of VIO methods where only estimated pose error and utilization of the computing resources were considered. However, the performance of them according to trajectory planning was not treated. Trajectory planning can be one of the main reasons for the drift of localization from VIO since the sensors acquire the data at their sampling rates. In this paper, we analyze the performance of VIO on various trajectories that have different orders of polynomials for the desired position over time and hence different velocity, acceleration, and the first, second derivative of acceleration, using simulator which reflects the real dynamics of UAV. As a result, the higher the order of polynomial for trajectory, the better performance of VIO can be observed.