Initialization-free monocular visual-inertial state estimation with application to autonomous MAVs

Abstract

The quest to build smaller, more agile micro aerial vehicles has led the research community to address cameras and Inertial Measurement Units (IMUs) as the primary sensors for state estimation and autonomy. In this paper we present a monocular visual-inertial system (VINS) for an autonomous quadrotor which relies only on an inexpensive off-the-shelf camera and IMU, and describe a robust state estimator which allows the robot to execute trajectories at 2 m/s with roll and pitch angles of 20 degrees, with accelerations over 4 m/s2. The main innovations in the paper are an approach to estimate the vehicle motion without initialization and a method to determine scale and metric state information without encountering any degeneracy in real time.