Attack and Defense on Aircraft Trajectory Prediction Algorithms
Quincy G. van Iersel, Alejandro Murrieta Mendoza, Roberto Salvador Félix Patrón, Seyed Mohammad Hashemi, Ruxandra Mihaela Botez · AIAA AVIATION 2022 Forum · 2022
View Video Presentation: https://doi.org/10.2514/6.2022-4027.vid The aviation industry needs lead to an increase in the number of aircraft and their flights. When the number of flights within an airspace increases, the chance of a mid-air collision (or collision) increases. Collision Avoidance Systems such as the Traffic Alert and Collision Avoidance System (TCAS) and Airborne Collision Avoidance System (ACAS) are currently used to alert pilots for potential mid-air collisions. The TCAS and the ACAS use algorithms to perform Aircraft Trajectory Predictions (ATPs) to detect potential conflicts between aircrafts. In this paper, three different aircraft trajectory prediction algorithms are discussed by using existing aircraft trajectory data containing multiple different aircraft types with different flight patterns. With this dataset, the future aircraft heading is predicted using the latitude, longitude, altitude, velocity and time. The three algorithms’ performances were evaluated in terms of their accuracy and robustness. These trajectory prediction algorithms were as well evaluated in the case of adversarial samples. Although algorithms can find reliable ATPs, earlier research has shown that they are also vulnerable against adversarial attacks that produce adversarial samples. Adversarial samples are similar to original samples from the dataset. These perturbations can cause trained algorithms to predict unreliable trajectories, which cause a security threat for learning-based trajectory algorithms, as adversarial attacks can result in intentional collisions. Adversarial training is applied as defense method in order to increase the robustness ATPs algorithms against the adversarial samples. The adversarial samples are included in the training data in order to make the algorithm more robust in the case of an adversarial attack. The findings in this research show that, comparing the three algorithm’s performance, the extreme gradient boosting algorithm is most robust against adversarial samples and adversarial training will benefit the robustness of the algorithms against lower intense adversarial samples. The contributions of this paper concern the evaluation of different aircraft trajectory prediction algorithms, the exploration of the effects of adversarial attacks, and mainly the effect of the defense against adversarial samples with low perturbation compared to no defense mechanism.