Unmanned Aerial Vehicles for Maintenance of Themed Environments

Abstract

Man-made infrastructure requires performing regular inspection and maintenance such that any structural damages or degradation can be resolved in a timely man- ner. Themed environments pose the need for cleaning and painting large and complex 3D structures on a regular basis in order to maintain their aesthetic ap- peal. Aerial robots could provide a faster and economical way to maintain large areas in a shorter time. This thesis aims to investigate the feasibility of using Unmanned Aerial Vehicles (UAVs) for performing such tasks. In particular, we aim to investigate the type of design, sensing and planning frameworks needed to enable UAVs to perform precise maintenance tasks in close proximity to complex structures. While we develop our robotic solution for themed environments, the so- lutions presented and findings are relevant in other areas of structural maintenance as well. The primary contribution of this thesis is the design of an aerial platform suit- able to perform large-scale painting on 3D surfaces. Developing such a solution de- manded solving several crucial elements. For a UAV to be able to paint a generic 3D surface, it needs to accurately adjust its location with respect to the target surface. Specifically, the UAV has to maintain a certain distance and orientation at each point. This requires very accurate perception of the UAV’s surroundings. Hence, our first contribution is the development of a graphics processing unit (GPU) accel- erated dense reconstruction and localization pipeline that runs completely onboard the UAV with limited computation capabilities. Our perception stack can run at 60Hz fully onboard. Along with the ability to accurately localize relative to the 3D surface, the proposed approach can generate sub-centimeter maps of the surface using a hybrid GPU-CPU framework. These maps form a crucial backbone for other parts of the thesis. The following part of the thesis focuses on the hardware design of the Paint- Copter UAV and several software components that allow automated spraying on a given surface. PaintCopter is designed as a tethered platform, with dedicated power and paint lines, allowing extended mission duration times. With the spray gun mounted on a Pan-Tilt Unit (PTU), targeted spraying on precise locations is achieved, despite the platform disturbances. We show the capability of Paint- Copter in being able to paint any desired linear pattern on generic 3D surfaces in a fully autonomous fashion. However, such an approach makes it extremely hard to accurately modify the surface’s appearance. Therefore, the next contribution is the development of a Virtual Reality (VR) technology based user interface for the PaintCopter UAV. This interface allows the user to immerse in a virtual environ- ment consisting of the 3D model of the surface. The user will be able to virtually paint at desired locations on the target surface using a virtual spray gun. The in- formation from the virtual environment is processed to execute a painting mission by the PaintCopter to generate a similar output as in the virtual environment. In this way, the painter will be able to paint the structure in a similar fashion as he/she would in the real-world. The final part of the thesis focuses on automating the process by performing end-to-end planning of a painting process. With this approach, the user provides the desired appearance of the structure in the form of a colored 3D model and our software plans the entire mission including what to paint and where to paint in order to achieve the desired appearance. Our data-driven approach is fully self-supervised and does not require any sort of intervention from a human expert. Through this thesis, we show that UAVs can be designed to perform accurate and fast cleaning/painting jobs on large 3D surfaces. We provide solutions to address the challenges involved in painting or cleaning such structures with high precision. We develop software that allows planning and executing (a) semi-autonomous mis- sions with human supervision, or, (b) fully autonomous end-to-end missions. The proposed solutions have been extensively tested on the field, over extended periods of time, validating the reliability and effectiveness of the system.