Strategic Mitigation Against Wireless Attacks on Autonomous Platoons

Abstract

With the increased demand for connected and autonomous vehicles, vehicle platoons will play a significant role in the near future, enhancing traffic efficiency and safety. However, their reliance on wireless communication channels makes such systems susceptible to a range of cyber-attacks. An intelligent adversary could target the platoon through message falsification between vehicles to carry out high-impact attacks. This would create persistent degradation of platoon stability or even cause catastrophic collisions. In this paper, we present a novel, end-to-end attack detection and mitigation framework. We use a deep neural network as an example anomaly detector tuned to reduce false alarm rate. We then model the interactions between the imperfect detector, the intelligent adversary and the defense system as a non-cooperative security game with imperfect information. In this setting, the adversary performs a test-time boiling frog attack against the detector. The Nash-equilibrium solution considers the downstream effects of the test-time attack, to guide the control system reconfiguration for the vehicles to mitigate communication-based attacks. The simulations conducted in a sophisticated simulator demonstrate the potential for real-world online deployment in a distributed manner. Results show that our approach outperforms baseline methods by up to $30\%$ in terms of increase of defense utilities, leading up to $176\%$ increase in minimum inter-vehicle distances for collision avoidance under attacks.

We gratefully acknowledge support from the DSTG Next Generation Technology Fund and CSIRO Data61 CRP ‘Adversarial Machine Learning for Cyber’, and a CSIRO Data61 PhD scholarship.