Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Navigation of drones is predominantly based on sensor fusion algorithms. Most of these algorithms make use of some form of Bayesian filtering with a majority employing an Extended Kalman Filter (EKF), wherein inertial measurements are fused with a Global Navigation Satellite System (GNSS), and other sensors, in a kinematic framework to yield a navigation solution (position, velocity, attitude, and time). However, the long-term accuracy of this solution is exacerbated during the absence of satellite positioning, especially for small drones with low-cost MEMS inertial sensors. On the other hand, a recently proposed vehicle dynamic model (VDM)-based navigation system has shown significant improvement in positioning accuracy during the absence of a satellite positioning solution, although in a mostly offline setting. In this article, we present the software architecture of its real-time implementation using Robot Operating System (ROS) that separates and interfaces its core from a particular hardware. The presented implementation asynchronously handles different sensor data in a modular fashion and allows i) adapting the underlying aerodynamic model, ii) including complementary sensors, and iii) reducing the dimensionality of the EKF state space at run-time without compromising the navigation performance. The real-time performance of the proposed software architecture is evaluated during long GNSS absences of up to eight minutes and compared to that of inertial coasting.
Jan Skaloud, Pasquale Longobardi
Giovanni De Cesare, Paolo Perona, Robin Schroff