Your behavior appears to be a little unusual. Please verify that you are not a bot.


Research Online: Robust tightly coupled GNSS/INS estimation for navigation

By Omar Garcia Crespillo, Daniel Medina, Anja Grosch, Jan Skaloud and Michael Meurer / Presented at the European Navigation Conference, Lausanne, Switzerland, May 2017


Simulation Comparison: Classical GNSS/INS EKF and robust Huber EKF. Click to enlarge.

We designed a tightly-coupled integration between GNSS and inertial navigation systems (INS) where we modify the update step of a classical Extended Kalman Filter (EKF) to consider different robust estimators (such as M-estimators). We consider different faulty scenarios where the pseudoranges contain one or several non-modeled biases. The tightly-coupled GNSS/INS robust Kalman filter performance in the presence of biases is compared with the classical EKF and with a loosely-coupled Robust-GNSS/INS approach. The robust tightly-coupled version is able to minimize more efficiently the biases effect thanks to the direct redundancy of the inertial sensor within the robust estimator.

We set a simulated scenario based on a realistic trajectory and generate both GNSS and inertial measurements following state-of-the-art error models. We analyze the filter behavior under the presence of pseudorange measurement faults. For that purpose, we have run 100 Monte Carlo simulations over the given trajectory, and we have generated synthetic pseudorange biases of 40 meters in satellites PRN 18 and PRN 24 every 20 seconds. The filter error performance in the position domain is shown for the classical EKF and for a robust EKF based on Huber estimation criteria, the mean simulation error as well as the 95 error confidence interval. The classical EKF is highly affected by the sudden biases, and their effect influences for some seconds the estimation, while the robust Huber EKF is less sensitive to the presence of these biases because it is able to better adjust the estimation to minimize their effect in the final position estimation error.