Despite numerous advantages cooperative perception offers, it is also prone to sensor faults and data manipulation, where malicious agents or sensors can intentionally or unintentionally share faulty information. As a result, vehicles might generate faulty perception of the traffic situation and make erroneous decisions, potentially compromising road safety. In this paper, we propose a decentralized algorithm that leverages consensus-based detection using the states and their covariances to identify manipulated data in collective perception messages (CPM). The algorithm focuses on detecting attacks on objects that can be observed by multiple vehicles by quantifying how aligned each observation pairs are. In a scenario involving three vehicles perceiving the environment, the algorithm detected 67.57% of 37 attacks, which is 13.5% higher true positive rate than the baseline model.
Consensus-based Attack Detection and Cooperative Perception of Connected Vehicles
2024-10-07
1284183 byte
Conference paper
Electronic Resource
English
COOPERATIVE PERCEPTION WITH DEEP REINFORCEMENT LEARNING FOR CONNECTED VEHICLES
British Library Conference Proceedings | 2020
|