Anomaly Detection in Intersection Control

Anomaly Detection in Intersection Control: Sliding Mode Observer Based Anomaly Detection in Virtual Platooning Enabled Intersection Control

Janse, Eline (TU Delft Mechanical, Maritime and Materials Engineering)

Ferrari, Riccardo M.G. (mentor)
Jarmolowitz, Fabian (mentor)
Keijzer, T. (mentor)
van Wingerden, J.W. (graduation committee)
Ferranti, L. (graduation committee)

Delft University of Technology

2020-07-06

Road intersections have a large impact on road accidents and travel delay. Applying infrastructure, such as stop signs and traffic lights, are supposed to prevent collisions on the intersection. However, such methods contribute to the travel delay, while accidents still occur due to human errors. Considering the rise of automation within vehicles, the automation of intersections becomes possible. Vehicles exchange information with each other (distributed techniques) or with a central unit (centralized techniques) to determine the velocity profile needed to efficiently cross the intersection without causing any collisions. This information exchange is established through the creation of spontaneous wireless networks between vehicles and infrastructure, also known as a Vehicular Ad-Hoc Network (VANET).

However, the addition of wireless communication to autonomous vehicles presents a dangerous vulnerability to cyber attacks. In the literature, many studies investigate what types of cyber attacks are most likely to occur on the VANET. From these studies, it is chosen to test the cyber attack called "false data injection" on Virtual Platooning based Intersection Control.

It is shown in simulations that an alteration in the broadcasted messages for a duration of $1.3$ seconds already leads to a collision on the intersection. In order to prevent such events from happening, a Sliding Mode Observer (SMO) is designed to monitor the received data from the surrounding vehicles. Treating the false injected data as an unknown input to the system, the SMO is designed to reconstruct the anomalous data.

The accuracy of the unknown input reconstruction using an SMO depends on how well the model describes the dynamics of the vehicles. Thus the SMO is designed for both a linear as well as a non-linear vehicle model. To analyze the performance of the SMO, the observer is applied to both simulations and experiments. Under the assumption that the vehicles can measure the relative velocity to each other, both the simulations and the experiments show promising results.

Intersection Control
Sliding Mode Observer
Anomaly Detection
Autonomous vehicles
Anki overdrive

http://resolver.tudelft.nl/uuid:9ce8cace-1824-4111-87f5-e45572771a3d

Student theses

master thesis

© 2020 Eline Janse