Securing Vehicle State Estimation for Trustworthy V2X Communications
MSc Thesis Proposal by:
Nizbath Ahsan
Date: 23rd June, 2026
Time: 1PM
Location: Essex Hall 122
Abstract:
Connected Autonomous Vehicles (CAVs) rely on the Kalman filter for real-time state estimation, fusing data from GPS, IMU, LiDAR, RADAR, and camera sensors to compute position, velocity, and heading. These estimates govern safety-critical behaviors such as lane-keeping, obstacle avoidance, and braking, and are broadcast to neighboring vehicles via V2X communication protocols within the Vehicular Ad-hoc Network (VANET), enabling cooperative driving functions. The Kalman filter, however, assumes all sensor inputs are authentic and processes them without verification. Adversarial attacks targeting GPS, the primary source of absolute position and velocity, corrupt the Kalman filter's state estimate before any correction is possible. This corrupted state propagates across the VANET through Basic Safety Messages (BSMs), potentially triggering unsafe responses in multiple vehicles simultaneously. While prior work has addressed attacks in multi-sensor fusion, GPS in isolation as a pre-filter intervention point remains unexplored. This thesis proposes an unsupervised anomaly detection framework that intercepts adversarial GPS inputs upstream of the Kalman filter, to identify anomalous sensor behavior. The framework is trained and validated across multiple real-world driving datasets to assess generalizability. By protecting GPS integrity at the sensor level, this work aims to preserve accurate state estimation and cooperative driving safety under adversarial conditions.
Thesis Committee:
Internal Reader#1: Dr. Muhammad Asaduzzaman
Internal Reader#2: Dr. Dan Wu
Advisor: Dr. Ikjot Saini
