In the new era of intelligent systems, small-scale electronics, and advanced sensors, the application use-cases and operational capabilities of autonomous robots are exponentially increasing. Through their ability to execute complex tasks while relying only on onboard sensors and computation, autonomous field robots are showing promising results in inspection of infrastructure, search-and-rescue or surveillance, maintenance tasks, or in general operations in areas that prove to be dangerous or hazardous for human operators to enter, while also often increasing the efficiency of such tasks. But, to enable robots to autonomously execute their missions the demands on onboard intelligence is increasing rapidly as well. As robot operations move into complex and dynamic environments, into mixed-traffic or multi-robot operational scenarios, or into missions that demand the exploration and navigation of completely unknown areas, a new paradigm of autonomous robot navigation and collision avoidance algorithms need to be developed as well. Towards achieving the vision of autonomous robots performing such tasks for the good of society, this new paradigm of navigation capabilities must first be extended to operate outside of simulation environments, and then to operations in realistic field conditions with all the challenges that comes with that. 

This thesis presents the development of a series of navigation methods for autonomous robots, with a specific focus on Unmanned Aerial Vehicles (UAVs). The vision of this thesis is to further the application areas of completely autonomous robotic platforms by extending their navigation capabilities: towards avoiding obstacles in their environment both static and dynamic, towards the critical perception-actuation link for reactive navigation, towards exploring and planning dynamic paths through previously unknown areas, and towards the coordination and safety in multi-agent robotic systems. This thesis also has a significant focus in the area of field robotics, meaning the ability to robustify and extend the robots onboard intelligence to handle the harsh conditions of real operations. This thesis will specifically investigate the application of autonomous UAVs in search-and-rescue tasks in subterranean environments, as well as a variety of inspection tasks in underground mines. In these environments the robots must operate completely autonomously without any assisting communication, computation, or perception infrastructure. In all of these areas a special focus has been placed on the real-life experimental validation of results and the required research to reach the readiness stage of such demonstrations, serving as the main motivator for the works presented in this manuscript.