UAVs operating in urban environments must manage the prevailing urban wind effects like gusts and local vortices that constitute a high potential risk for failure and safety. The present research project focuses on how to make an UAV flight in cluttered urban wind conditions more reliable and more efficient using a machine learning technique. A trajectory planning algorithm based on the traditional A* formulation was designed to determine the minimum energy path from a start to a final location taking into account the prevailing wind conditions. In order to obtain average wind conditions in an urban environment, full-scale CFD simulations were performed using OpenFoam® for various inlet wind directions. The computational model represents complex buildings of the campus of the Technical University of Berlin. The full-scale numerical simulations require high computational costs that make on-line calculations not feasible.
Hence, the key ingredient of the present investigation is the on-line reconstruction of the complete urban wind field using a Gappy POD approach with measurements from sparsely distributed sensors. Specifically, a full database (FOM) of the urban wind field is first simulated off-line using OpenFoam®. Then, a reduced-order estimate (ROM) of the wind field is obtained using sparse sensor measurements by solving a linear combination of the POD modes that are previously calculated on the full database during the off-line stage. Finally, the trajectory planning algorithm is applied to the reconstructed flow field and validated by comparison with the full-order model obtained by CFD. The energy consumption of an UAV was used in order to assess the accuracy of the wind field reconstruction using the Gappy POD approach to calculate a reliable trajectory. The energy consumption of the shortest path is compared to the energy consumption of