Development, integration and testing of algorithms to support autonomous flight, in the absence of GPS signal

Modern navigational systems should function properly and dependably not just when a GPS signal is present, but even when it is absent or maliciously blocked. Traditional navigation systems fail to work in many GPS-restricted settings, such as inside, caverns, canyons, or when GPS is jammed or not even available such as an outer planet scenario. Many researchers are now proposing a variety of methods to address these constraints. Visual-Inertial Odometry (VIO) is one of several approaches for dealing with GPS-denied navigation that has piqued the scientific community's curiosity. Only a tiny portion of the offered methods can produce desirable accurate results and be considered for applications where acceptable Size, Weight, and Power (SWaP) are restricted, due to large processing needs and insufficient resilience when addressing complicated real-life scenarios. ?? The purpose of this work is firstly to provide a concise but complete classification of VIO algorithms and to offer a panoramic on State of the Art Techniques for UAVs Navigation in Critical environments. A deep analysis is carried out on the OpenVINS framework, published by the Robotics Group of Delaware's University with the purpose of enabling and facilitate the development and evaluation of new VIO algorithms. ?? The thesis focuses then to provide an embedded platform with enough computational capacity, based on Rock Pi N10 by Rockchip, to run the VIO exploiting both hardware and software needs to execute in real-time the pose estimation. Since it is crucial in order to obtain robust estimations, a complete walk through for sensor calibration with Kalibr and Kalibr\_allans provided. ?? The system is finally evaluated inside a martian-like environment, thanks to the collaboration with Thales Alenia Space Italia, exploiting the framework capabilities and integrated analysis tools.