Combined RGBD-inertial based state estimation for MAV in GPS-denied indoor environments

This paper presents a integrated navigation approach for state estimation of a micro aerial vehicle (MAV) that is capable of autonomous flight in GPS-denied, indoor environments. The solution combines RGB-D sensor and inertial sensors in a tight-coupling navigation manner. Motion estimates from RGB-D visual odometry and inertial measurements are fused using an improved Extended Kalman Filter-based fusion algorithm to provide an accurate estimate of the relative position, velocity and attitude. Instead of using a global reference frame, a view-based map is employed and the algorithm maintains the position and heading relative to the current map node in the fusion algorithm. In addition, a closed-form covariance is developed to qualify the uncertainty of the RGBD visual odometry measurements, which is utilized for state update of the navigation filter. Our approach allows efficient measurement updates and enables the incorporation of RGBD visual odometry uncertainty. Experimental results of a quadrotor MAV flying in a GPS-denied indoor environment demonstrate the performance of the proposed approach. Comparisons of state estimates with ground truth measurements are also provided.