Pre-integrated dynamics factors and a dynamical agile visual-inertial dataset for UAV perception

Abstract

For the past few years, the rapid development of unmanned aerial vehicle (UAV) technology has been met by increased interest in these platforms for a wide range of applications. Particularly, the autonomous navigation of these vehicles is of great interest for applications such as surveillance, mapping, searching, agriculture, and film-making, to name a few. But autonomous UAV research has a long way to go to meet the capabilities and robustness required in many of these applications. This work presents two contributions towards closing that gap: pre-integrated dynamics factors for factor graph visual-inertial odometry (VIO), and a large-scale dataset with a great variety of visual, inertial, and dynamical sensor data from a quadrotor platform. The pre-integrated dynamics factors were tested on a challenging subset of the dataset and showed an improvement in robustness of a VIO system. The size and variety of the dataset make it a valuable tool for evaluating and testing visual-inertial estimation algorithms, as shown with the dynamics factors. Both contributions facilitate the development of more robust autonomous UAV navigation systems.