Rotary-wing micro aerial vehicles (MAVs) are disrupting geomatics, logistics, and maintenance industries. Being airborne and easy to deploy, they are a great tool to investigate large areas, inaccessible structures, and dangerous environments. Commercial platforms exist that automatically generate digital maps of cities, count warehouse inventory or perform non-destructive testing on industrial assets. Many more applications are envisioned, but to turn those visions into reality, MAVs will have

more »

... to fly closer to structures, physically engage with the environment, and integrate novel sensor modalities while providing ever more autonomy to make these advanced functionalities available to non-expert users. This thesis presents three precision-critical applications that contribute to the state of the art of industrial aerial robotics. The first contribution is a multi-agent aerial robotic system to search, pick-up and relocate metallic objects. To interact with small, partly moving objects, the aerial robot requires precise navigation and detection capabilities as well as a compliant grasping process. Our system combines (i) GNSS empowered visual-inertial state estimation with (ii) collision avoiding, model predictive position control and (iii) geometric computer vision into a precise autonomous transportation system. The system was deployed in various environments, including successful participation in the Mohamed Bin Zayed International Robotics Challenge 2017. It shows that basic autonomous aerial physical interaction in outdoor environments is possible given welldefined task and environment constraints such as known target properties and workspace. The second contribution is an MAV equipped with a ground-penetrating synthetic aperture radar (GPSAR) for humanitarian landmine detection. Airborne GPSAR is highly dependent on the flight path and the precision with which the radar antenna positions are determined. In this work we present a navigation framework that allows generating arbitrary circular and stripmap GPSAR missions controlled at low altitude above ground level. A self-calibrating, factor graphbased localization framework combines dual receiver RTK GNSS with inertial measurements to estimate the position of the radar antennas during flight. A custom hardware triggering mechanism ensures temporal correlation of the navigation sensors reaching sub-µs accuracy with respect to GNSS time. The system is self-contained and enables autonomous optical and radar surveys. In various experiments we show the advantages of the custom system design, including uniform radar sampling, self-calibration, and localization batch optimization. Finally, we validate mapping of buried objects to demonstrate the system's suitability for humanitarian landmine detection. The final contribution is an automatic coverage path planner to enable aerial surveys with nadir imaging sensors in obstructed environments. The algorithm is based on exact cellular decomposition, which splits an admissible polygon flight area into simple polygons that can be covered efficiently with consecutive back-and-forth motions. Our algorithm improves the connection of the individual polygon coverage patterns by considering multiple starting points for each polygon. Formulating the problem as an equality generalized traveling salesman proi Abstract blem and basing it on a strong computational geometry foundation created an implementation that plans optimal coverage missions within seconds. The open-source software is popular in the robotics community and has been a valuable mission planning tool for our GPSAR missions. Overall this thesis focuses on high-precision aerial robotic state estimation and MAV mission planning under spatial and motion constraints. All contributions have been tested in selfcontained, real-world applications. The underlying software is available open-source to help bring forth a new generation of industrial aerial robots that operate autonomously in the vicinity of obstacles. Out of my eight fantastic years at ETH Zürich the five years at Autonomous Systems Lab (ASL) have definitely shaped me most both personally and professionally. This is mainly due to the unique atmosphere that Prof. Dr. Roland Siegwart creates. Roland, thank you for the infinite trust you give and the endless inspiration you seed. On my first day at the lab, which was the amazing 20 year anniversary retreat, I felt home immediately. This gave me full confidence to tackle any challenge that would occur in the course of my PhD. Your great reputation is certainly also a reason why today I can thank my co-examiners for accepting to co-supervise and examine this thesis. Thank you Prof. Dr. Jonathan Kelly and Prof. Dr. Martin Saska. Your work on sensor fusion and aerial robotic systems has always impressed me and has left its mark on my work. I hope this thesis will ignite some inspiration in return. An equally big thank you goes to my additional supervisors during my PhD, Dr. Nicholas Lawrance, Dr. Jen Jen Chung, and Dr. Juan Nieto. Thank you very much for the amount of time you invested in me to convert ideas into contributions, cover my back in field tests and teach me the basics of professional research. You have always been asking the right questions about my work, you helped to improve my writing substantially, and finally, you provided the guidance and orientation that eventually made me become a researcher. Unfortunately, the length of this section does not allow me to thank everybody from the ASL family, even though everybody deserves to be mentioned. You make this place so pleasant to be around. Over the years I did probably have a key experience with each one of you: Mark Pfeiffer, Sebastian Verling, Raghav Khanna, Matthias Gehrig, I would go surfing with you any time again. Thomas James Stastny, I think I know now why you should not solve every problem with MPC. If you also want to know, you have to read this thesis to the end. Same goes for Hannes Sommer who will probably find part of his knowledge inside this manuscript. Marcin Dymczyk thanks for welding my bike frame so I can get to work on time. Fadri Furrer, you made it very easy for me to integrate in the group when we first met on Lake Lucern. . . I could go on, but instead I want to highlight the collaboration I was able to enjoy. Luciana Borsatti and Cornelia Della Casa, no matter if you want to ship a hundred thousand worth of equipment, organize a conference or get a project going, you are always the first contact. Michael Riner-Kuhn, Matthias Müller and Markus Bühler, your support made my life so much easier, both teaching me electronics and building the hardware. Florian Braun and Lucas Streichenberg, you two have been absolute keystones in the development of the landmine detecting MAV. Thanks so much! In general, I want to thank the whole rotarywing group that formed my skill set and identity at the lab. You make this place extremely enjoyable and educational. Thanks to the old crew: