In recent years, research on aerial robots for physical interaction tasks has expanded rapidly, with a particular focus on multi-rotor aerial vehicles (MAVs). These platforms offer distinct advantages by providing fast access to high-altitude or hard-to-reach workspaces, making them valuable for inspection and maintenance applications where they can significantly reduce time, cost, and risks to human workers. Such applications, however, demand precise and stable physical interaction with the environment, driving ongoing advancements in MAV design and physical interaction control architectures.
Among these, omnidirectional MAVs, which can maintain arbitrary orientations while hovering, have attracted special attention for their exceptional dexterity. They enable operations that demand continuous reorientation of the tool, such as sliding along a pipe’s circumference for inspection. In contrast, conventional fully-actuated MAVs require an additional robotic arm to achieve such capability, increasing mechanical complexity, weight, and maintenance needs. Omnidirectional MAVs therefore offer a mechanically simpler and potentially more efficient alternative.
This thesis focuses on the development of a Stable and Reliable Physical Interaction Operation for Omnidirectional MAVs: The Omni-rotor. Robust physical interaction control for fully-actuated MAVs with coplanar rotors has already been demonstrated in the literature. Here, it is investigated whether these controllers can be effectively extended to non-coplanar omnidirectional designs, such as the Omni-rotor. The experimental validation is designed with a particular emphasis on physical inspection tasks.
An extensive set of experiments demonstrates the Omni-rotor’s ability to perform high-quality physical interactions while fully exploiting its omnidirectional capabilities. These experiments include point contact, sliding, and peg-in-hole tasks. To quantify performance, a dedicated set of performance metrics was determined based on relevant literature and the requirements of nondestructive testing (NDT). From a control and system integration perspective, this work developed a GenOM3 -based software architecture implementing an admittance-filter-based interaction controller. Additionally, a novel input allocation scheme was proposed and validated experimentally to mitigate the effects of modeling errors in the wrench generation of the propellers, thereby enhancing the accuracy of wrench estimation and improving physical interaction performance which is validated through a physical interaction experiment.