Filtering algorithms and avionics systems for unmanned aerial vehicles

This thesis deals with theoretical and experimental results of original filtering algorithms for unmanned aerial vehicles. These filters bypass the limitations of the commonly used estimators (e.g. the Extended Kalman Filter or the particular filters), and they can be easily and efficiently implemented in a low-cost avionics system. First, we propose "generic" invariant observers, which preserve the natural symmetries of the considered physical system. These observers merge measurements from standard low-cost sensors (such as inertial and magnetic sensors, GPS or barometers) to accurately estimate the state of the vehicle (attitude and heading, velocity and position) without any knowledge of the system dynamics. They possess a large domain of convergence; they are also easy to tune and computationally very thrifty. Then, we develop "specific" observers, well-designed for a given kind of aerial vehicle, which is in our case a mini quadrotor. Based on a physical model that takes into account the rotor drag, the proposed filters estimate the quadrotor velocity only from inertial measurements. Thus, the filtering algorithm leads to a velocity-controlled vehicle. We validate this approach through autonomous flights. Finally, we present in detail the integration of the low-cost avionics system: it is made of low-cost sensors and a microcontroller, on which the previous observers have been implemented. We validate the filtering algorithms through experimental comparisons with an expensive commercial device. By doing so, we highlight the excellent quality-price ratio of the filters developped in this thesis.