Thomas Konrad and Dirk Abel, RWTH Aachen University, Germany

View Abstract Sign in for premium content


Autonomous unmanned aerial vehicles (UAV) require three-dimensional state estimation with high accuracy, availability, integrity, and bandwidth. Both loosely and tightly coupled GNSS/INS navigation filters can provide such state estimates in real-time, each of them having practical drawbacks. On one hand, carrier-phase aided methods like Real-Time Kinematic (RTK) can enable loosely coupling with accuracies in the centimeter level. On the other hand, a stable phase lock can usually not be guaranteed due to the dynamic movement of UAVs, forcing a frequent fall back to RTK with floating ambiguities or even to pseudorange-based differential. A change from RTK to differential and vice-versa causes jumps in the estimated position, leading to unsteady flight behavior. Contrarily, tightly coupled filtering using single or double-differenced pseudoranges provides smooth estimates, but at the cost of higher expected mean errors of up to 1 m. In this paper, we present a method to improve the mean accuracy of a tightly coupled GNSS/INS navigation filter by means of a two-stage approach using RTK. Instead of directly performing pseudorange error corrections using differential techniques, we propose a second Kalman filter that estimates per-satellite pseudorange offsets in real-time using an RTK position when available. In short-term absence of RTK, we can maintain the current positional accuracy by means of propagating these errors in time. The method is implemented using Simulink, evaluated using real-time data from a multirotor flight and compared to the results of pure loosely and tightly coupled filtering. We show that our approach avoids sudden changes in position estimates, thus keeping the smoothness of the tightly coupled integration, while drastically reducing the expected mean position error to around 35 cm. The method is well suited to other applications, in which a smooth estimate is more vital than the respective best estimate at each individual time.