Robust Detection Framework for Adversarial Threats in Autonomous Vehicle Platooning

Stephanie Ness^1,2*

• ¹University of Vienna, Vienna, Austria
• ²Diplomatic Academy of Vienna, Vienna, Vienna, Austria

Autonomous vehicle platooning (AVP) has been introduced as an innovative approach towards enhancing vehicle dynamics, where several self-driving cars are grouped to travel in proximity to improve factors such as fuel consumption, safety, and traffic flow. However, the system is designed based on V2V and V2I communication to bring adversarial threats to the forefront. Such attacks may tamper with the information from the sensors, the control signals or decision-making mechanisms and bring about hazardous occurrences such as accidents or traffic jams. This paper discusses a new method to identify adversarial threats in AVP systems, combining active learning and machine learning classifiers. RF, GB, XGB, KNN, LR, and AdaBoost classifiers are used in the proposed method to improve detection accuracy and decrease the need for labelled data. The experimental outcome shows that RF with active learning provides the best accuracy of 83.91% for better identification of adversarial threats. The elaborated concept presents a probable answer to protect AVP systems and enable them to function stably and securely in conditions unidentified in laboratory environments.