Compromised ACC vehicles can degrade current mixed-autonomy traffic performance while remaining stealthy against detection

George Gunter, Huichen Li, Avesta Hojjati, Matthew Nice, Matthew R. Bunting, Carl A. Gunter, Bo Li, Jonathan Sprinkle, Daniel B. WorkarXiv 2021 · arXiv

Abstract

As semi-automated features like Adaptive Cruise Control (ACC) become standard, highways are transitioning into "mixed-autonomy" environments. This paper explores a subtle, systemic threat: traffic flow degradation via "Random Deceleration Attacks" (RDAs). We show that as few as 1% of compromised ACC vehicles can trigger large-scale congestion waves. These attacks are remarkably difficult to detect using current GPS-based anomaly detection or in-vehicle CAN bus monitoring because the resulting traffic patterns mimic natural congestion. Simulations on a 3km freeway stretch reveal a 7% reduction in mean speeds and significant economic costs. Hardware validation on a consumer vehicle confirms the physical feasibility of the attack.

Introduction

While most automotive cybersecurity research focuses on high-profile "kinetic" attacks, such as forcing a collision, this study investigates the systemic impact of subtle software compromises. The authors explore whether a supply-chain compromise of ACC software could be used to intentionally create "traveling waves"—stop-and-go ripples that degrade freeway performance. Distinguishing a malicious deceleration from a normal reaction to traffic fluctuations remains a significant challenge for current defense systems.

Methodology

The researchers employed a multi-faceted approach to evaluate the RDA threat:

  1. Attack Model: The RDA involves a compromised ACC unit occasionally overwriting its intended acceleration with a subtle braking command.
  2. Simulation (Anti-Flow): Using the Anti-Flow framework (built on SUMO and Flow), the team simulated a 3km stretch of a US freeway. They used microscopic traffic models where individual vehicle behaviors are governed by ordinary differential equations (ODEs).
  3. Defense Evaluation: An anomaly detector using a CNN + LSTM autoencoder was developed to analyze GPS traces, attempting to flag malicious vehicles based on speed profile reconstruction errors.
  4. Hardware Validation: A Toyota RAV4 was retrofitted with a CAN bus injection tool to demonstrate that ACC commands could be successfully overwritten in a real-world setting.
The attack model involves compromising the ACC ECU to inject malicious deceleration commands onto the CAN bus.

The study utilized high-fidelity simulated freeway data, GPS traces for training detection models, and real-world CAN bus logs from experimental RDA executions.

Key Insights: The Stealth of the "Victim Effect"

A critical finding is the victim effect: when an attacker slows down, the vehicles behind it (the victims) must also decelerate. To a GPS-based detector, these victims often appear just as "anomalous" as the attacker, making isolation nearly impossible.

Furthermore, the attack exploits a gap in current CAN bus defenses. Traditional tools look for unusual message frequencies or IDs. However, the RDA simply alters the data within perfectly valid, expected messages, rendering standard in-vehicle security blind to the intrusion.

Results and Impact

The simulations demonstrated significant degradation of traffic performance:

  • Speed and Throughput: In moderate congestion, mean commuter speeds dropped by over 7%. In high-density scenarios, network throughput decreased by 3%, or approximately 300 fewer vehicles per hour.
  • Economic Cost: The delays were estimated to cost over $300 USD per kilometer-hour for the modeled stretch.
  • Detection Failure: The deep learning anomaly detector proved ineffective, with a True Positive Rate (TPR) below 3% in most scenarios.
Heatmaps showing the propagation of congestion waves (traveling waves) triggered by a single perturbation in dense and light traffic scenarios.

Hardware tests confirmed that a real vehicle could be forced to decelerate at -1.7 $m/s^2$ for 10 seconds, proving the RDA is physically possible on existing consumer hardware.

Conclusion and Future Work

The study highlights a systemic vulnerability in mixed-autonomy traffic. Even a 1% compromise rate can lead to significant infrastructure-level disruption. Future research should address the role of human intervention—whether drivers notice and override these decelerations—and the potential for more advanced monitoring systems, such as overhead cameras or 100% vehicle tracking, to mitigate these stealthy threats.

Summary updated

Abstract

We demonstrate that a supply-chain level compromise of the adaptive cruise control (ACC) capability on equipped vehicles can be used to significantly degrade system level performance of current day mixed-autonomy freeway networks. Via a simple threat model which causes random deceleration attacks (RDAs), compromised vehicles create congestion waves in the traffic that decrease average speed and network throughput. We use a detailed and realistic traffic simulation environment to quantify the impacts of the attack on a model of a real high-volume freeway in the United States. We find that the effect of the attack depends both on the level of underlying traffic congestion, and what percentage of ACC vehicles can be compromised. In moderate congestion regimes the attack can degrade mean commuter speed by over 7%. In high density regimes overall network throughput can be reduced by up to 3%. And, in moderate to high congestion regimes, it can cost commuters on the network over 300 USD/km hr. All of these results motivate that the proposed attack is able to significantly degrade performance of the traffic network. We also develop an anomaly detection technique that uses GPS traces on vehicles to identify malicious/compromised vehicles. We employ this technique on data from the simulation experiments and find that it is unable to identify compromised ACCs compared to benign/normal drivers. That is, these attacks are stealthy to detection. Stronger attacks can be accurately labeled as malicious, motivating that there is a limit to how impactful attacks can be before they are no longer stealthy. Finally, we experimentally execute the attack on a real and commercially available ACC vehicle, demonstrating the possible real world feasibility of an RDA.
BibTeX
@misc{gunter2021compromised,
  author = {George Gunter and Huichen Li and Avesta Hojjati and Matthew Nice and Matthew R. Bunting and Carl A. Gunter and Bo Li and Jonathan Sprinkle and Daniel B. Work},
  title = {{Compromised ACC vehicles can degrade current mixed-autonomy traffic performance while remaining stealthy against detection}},
  year = {2021},
  eprint = {2112.11986},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2112.11986},
}
Figure