Structural Inspection is a key part of the maintenance and upkeep life-cycle for every piece of infrastructure we have built and will build. However it still remains diﬃcult, expensive, time consuming and, above all, dangerous for engineers to inspect and maintain many of the larger and taller structures out there. With the rise in consumer electronics leading the development of cheap and low cost Unmanned Aerial Vehicles (UAVs), these devices have the potential to make structural inspection not only safer, but more eﬃcient and cheaper. We believe that a multi-agent solution can provide the user with a more eﬃcient and crucially, a more robust method for structural inspection compared to the single-agent methods used today. With this project, we have sought to identify and understand the main features of the robust structural inspection problem so that we can develop a solution which can be applied to a real life scenario. We have developed and validated a novel evaluation metric for a speciﬁc class of coverage solutions which aim to divide up a single cyclical tour of the environment into multiple static, non-reactive, agent trajectories, one for each agent. We then question the static assumption of the previous method and conduct a brief exploration of the use of dynamic re-routing methods. Finally, we build a new simulation environment with GPU acceleration based on the AVSCPP planner using C++, OpenGL, PCL and Octomap libraries, upon which we applied some of the methods we discuss with the aim of observing the eﬀect of true spatial location on these coverage methods, as well as provide a foundation for future implementations.
Masters of Research Thesis