Cybersecurity Analysis in the UAV Domain: the Practical Approach of the Labyrinth Project

Abstract

In the last decades, Unmanned Aerial Vehicles (UAVs) are finding more and more fields of application. Their flexibility and cost-efficiency make them useful to support complex operations in agriculture, remote sensing or construction, just to name a few. In the Labyrinth project we aim at investigating the applicability of UAV usage to critical scenarios like air, water and road traffic control or emergency, with a strict focus on safety, security and efficiency. This involves also the cybersecurity aspect, which is the main focus of this work. UAVs used in critical applications are in fact potentially exposed to a wide set of cyber threats. The NIST cybersecurity framework [17] defines five different security functions which are: identify, protect, detect, respond and recover. In this paper we address the identify and detect functions with an approach involving threat analysis and anomaly detection. Firstly, we identify which threats pose a significant risk to the Labyrinth use case, for instance leading to the collision of UAVs in case an attacker is successful. Secondly, we present a machine learning-based pipeline aimed at detecting deviations in the position reportings of the drone, to support the detect function during flight operations. The pipeline is tailored to the Labyrinth system reporting needs and is based on unsupervised machine learning to overcome the lack of labeled data. Anomalous points, i.e., points deviating from a coherent path, potentially because of a cyber-attack or a failure, are visually separated from the coherent ones and marked as noise. To prove its robustness, we test the pipeline introducing artificial perturbations in the data.