ATTACKS ON SENSORS FOR AUTONOMOUS VEHICLES: DETECTION AND ISOLATION

Abstract

This research looks on the cyber security threat posed by sensor assaults to autonomous cars. Specifically, a
model-based strategy is proposed to allow safe localization of autonomous vehicles by identifying the origins of sensor
assaults. As a countermeasure against cyberattacks, we propose sensor redundant work, or the use of many sensors to
monitor the vehicle's position in real time. Banks of attack detectors are constructed using a Kalman filter with an extension
(EKF) and an overall sum (CUSUM) distinction to identify outliers in sensor readings. The recursive EKFs are used
CUSUM discriminators are built to analyse the remaining energy produced by their combined EKF in order to spot any
deviation of the sensor's output from the projected ask derived from a numerical model of the vehicle, allowing for more
accurate position and orientation estimates. To keep an eye on the discrepancy between various sensors' readings, a new type
of detector is devised that combines information from many sensors. On the basis of these rules, an isolation strategy is
developed to identify the offending sensor. results from each and every detector. Our proposed framework has been shown
to be useful based on data from actual automobiles.