On-demand Landmark Activation to aid Navigation for Advanced Air Mobility

Navigation requirements for Uncrewed Aerial Systems (UAS) utilizing the Advanced Air Mobility (AAM) framework requires complete or partial autonomy that currently relies primarily on Global Positioning System (GPS) signals. However, with AAM services extending to areas like urban canyons that shows considerable degradation in the quality and accuracy of GPS data, alternate means of navigation are being explored. This paper proposes one such method by injecting synthetic known features as landmarks into the environment, that can be activated \textit{on demand} by the vehicles. Accurate navigation requires precise localization (knowledge of the UAS' current location with respect to a common reference frame or map) which can be achieved by measuring range to these landmarks with the help of onboard sensors. Practical considerations like cost of operation, communication cost, bandwidth to name a few prevents placing these landmarks abundantly in the environment. Hence, this paper explores the option of On-Demand Landmark Activation using localization uncertainty as a constraint in the system. The frequency of measurements received from landmarks and/or from neighboring vehicles in a cooperative setting is used as a metric in developing a data-driven approach for modeling uncertainty prediction. MATLAB is used to generate the training and test datasets for single and cooperative UAS cases which shows that Gaussian Process Regression is a viable technique for generating a closed-loop uncertainty model. An Unreal Engine (UE) based generalized high-fidelity simulator in AirSim has also been developed for a cooperative navigation scenario as part of this paper to provide datasets for robust testing in different environments.