Invariant Extended Kalman Filter-based state estimation for MAV in GPS-denied environments

This paper presents a RGB-D aided inertial navigation system that uses RGB-D sensor and low cost inertial measurement sensors (IMU) to provide state estimates for micro aerial vehicles (MAV) in GPS-denied indoor environments. The state estimation approach is based on the invariant observer theory which is developed for systems possessing symmetries. In our system, we review the invariant observer theory and design the invariant observer (Invariant Extended Kalman Filter, IEKF) based on the analysis of system symmetry for the RGB-D aided inertial navigation model evolving on a Lie group. In addition, a robust RGB-D based motion estimation approach is developed to provide relative pose estimates using feature correspondences captured by the RGB-D sensor. The RGB-D estimates are fused with inertial measurements through the IEKF-based observer which yields a simplified error dynamics and simplifies the calculation of gain matrices. The resulting framework is implemented and validated on a MAV, and experimental results from actual indoor flight tests demonstrate the effectiveness of the approach.