Search for: All records

As city populations continue to rise, urban air mobility (UAM) seeks to provide much needed relief from traffic congestion. UAM is enforced by electrical vertical takeoff and landing (eVTOL) vehicles, which operate out of a vertiport, akin to the relationship between planes and airports. The vertiport has an air traffic controller (ATC) tasked with managing each eVTOL, ensuring they reach their destinations on time and safely. This task allocation problem can be difficult due to inadvertent issues such as mechanical failure, inclement weather, collisions, among other uncertainties that may arise. This paper provides a novel solution to this Urban Air Mobility - Vertiport Schedule Management (UAM-VSM) problem through the utilization of graph convolutional networks (GCNs). GCNs allow us to add abstractions of the vertiport space and eVTOL space as graphs, and aggregate information for a centralized ATC agent to help generalize the environment. We use Unreal Engine combined with Airsim for high fidelity simulation. The proposed GRL agent will be trained in an environment without extra uncertainties and then tested with and without those uncertainties. The performance will be examined side by side with a random and first come first serve (FCFS) baseline.