Identification and Integration of Aerodynamics into Fixed-Wing Drone Navigation

Autonomous navigation of drones in environments with intermittently poor reception of Global Navigation Satellite System (GNSS)-observables remains a complex problem when using small, low-cost inertial sensors. Conventional kinematic frameworks, fusing Inertial Measurement Unit (IMU) data with GNSS, often falter under such conditions, yielding a rapidly exacerbating positioning accuracy that does not allow practical utility. To that effect, a Vehicle Dynamic Model (VDM)-based navigation system has shown dramatic improvement in the drone's positioning by incorporating a model of its aerodynamics into the sensor fusion architecture. Such an approach, however, relies on the knowledge of model parameters that are specific to the operating vehicle, deduced primarily through wind tunnel experiments and/or fluid dynamics simulation. In this thesis, we present a novel algorithm, utilizing recorded flight data exclusively, to identify the values of unknown model parameters of two geometrically different fixed-wing drones - one being conventional, and the other a delta-wing. Our approach proves to be independent of the choice of the winged platform and prior numerical knowledge of its aerodynamic coefficients, apart from being cost-effective and considerably quick to implement. Subsequently, we compare the performance of the parameters identified by the approach above, to those obtained from wind tunnel experiments for a delta-wing drone, in a VDM-based framework, finding the employment of both parameter sets to be nearly comparable. Owing to the availability of an aerodynamic model and numerical values of model parameters, we thereafter develop a software architecture for the real-time implementation of the VDM-based system using Robot Operating System (ROS), in a way that separates and interfaces its core from a particular hardware and choice of the drone. Next, we extend this software architecture to accommodate multiple IMUs, in contrast to existing studies that have employed only one. As a minor sidestep from our study of navigating drones in GNSS-denied scenarios, we demonstrate the utility of the VDM-based system as a synthetic wind sensor during the availability of GNSS. In this aspect, our work stands out against most of the existing studies, for fixed-wing drones, that have either used an airspeed/other sensor or relied on wind-tunnel calibration for wind estimation. In tandem with the above distinct contributions, we propose a novel linear estimation methodology to deduce deterministic accelerometer errors, such as switch-on bias, without any initial guess and multi-iteration requirements that are prevalent in most existing studies. This methodology has also been embedded into a ROS-based framework and can feed calibrated accelerometer data to any navigation system, conventional/VDM, or others prior to take-off, indirectly improving the quality of initial orientation. All aforementioned tasks are validated in several field tests across 3 fixed-w