Indoor quadrotor state estimation using visual markers

Abstract

This paper discusses the problem of estimating the full state-vector (position/orientation) of an AR.Drone quadrotor using measurements from an inertial measurement unit (IMU) and an on-board camera taking images of predefined markers. The platform used is an inexpensive commercial quadrotor. The open-source Robot Operating System (ROS) is used to manage communication with the quadrotor. To estimate the AR.Drone states, an extended Kalman filter is used. The state estimates are propagated using a nonlinear dynamic model of the AR.Drone available in the literature. The estimation error covariance is propagated through the continuous-time Riccati equation using the model Jacobian. The estimated states are updated based on measurements of angular velocity from the IMU along with position and orientation from the camera. Convincing experimental results are presented. The work introduced here allows for an overall inexpensive setup for estimating the states of a quadrotor for flight in GPS denied environments using visual markers.