Camera-aided Inertial Navigation for Unmanned Aerial Vehicles using Non-linear Observers
Abstract
The objective of this thesis is to investigate solutions for a robust autonomous navigation system for a fixed-wing UAV. Robustness requires the ability to detect faults in the system and to compensate for the consequent deviations from the nominal behaviour. The specific aspect studied here is sensor redundancy, which is the inclusion of additional sensors whose measurements can constitute backup information in case some other measurement becomes unreliable.
The focus is on inertial navigation of a fixed-wing unmanned aerial vehicle using low-cost micro-electro-mechanical-system (MEMS) inertial sensors and machine vision. The information coming from the sensors is fused by means of stand-alone non-linear observers and combinations of non-linear observers and the linearized Kalman filter. The observer designs are based on a previous work that combined an attitude observer with a translational motion observer using a feedback interconnection, which was proved to be globally exponentially stable. Stability proofs are given for some of the proposed observers; all observers are tested on experimental data.
Machine vision is used to calculate first the optical flow from a series of images, and second the body-fixed velocity of the UAV. Two different methods for obtaining the velocity from the optical flow are investigated. With respect to previous works machine vision can replace the magnetometer, a sensor that suffers from various disturbances when surrounded by metal and electric circuits. The estimation of position and linear velocity is aided by a global navigation satellite system, but when this is not available it is necessary to rely on dead reckoning. This thesis also includes contributions towards this aspect of navigation, proving that the velocity calculated through machine vision and non-linear observers helps reduce the growth rate of the dead reckoning position error.
Non-linear observers and the non-linear extensions of the Kalman filter have advantages and disadvantages. A recent state estimator, referred to as exogenous Kalman filter, is formed as a cascade of a non-linear observer and a linearized Kalman filter, and is able to retain the advantages of each component while discarding their shortcomings. In this thesis it is tested for the first time on experimental data and compared with a stand-alone non-linear observer and an extended Kalman filter, providing further insight into this novel state estimation algorithm.
The work is organized as a compendium of scientific papers. It includes 3 conference papers, 2 journal papers, and 1 book chapter. Two additional papers are mentioned, but are outside the scope of this thesis.