A Secure Sensor Fusion Framework for Connected and Automated Vehicles Under Sensor Attacks

Chen Lv (吕辰), Tianci Yang, Chen Lv

This paper, published in the Journal of LaTeX Class Files (and archived on arXiv), presents a robust defense strategy for Connected and Automated Vehicles (CAVs) facing malicious sensor attacks. As we move toward platoons of self-driving cars communicating over wireless networks, the "attack surface" grows; a single spoofed LiDAR or Radar could lead to catastrophic collisions.

The authors, Tianci Yang and Chen Lv, propose a framework that doesn't just detect attacks but maintains safe operation even while they are happening.

TL;DR

The researchers developed a secure sensor fusion algorithm that provides a reliable estimate of vehicle data even when nearly half of the sensors are compromised by cyberattacks. By combining this "attack-resilient" estimate with a specialized $H_\infty$ controller, they demonstrate that a vehicle platoon can maintain stability and safety despite active, time-varying sensor spoofing.

What Problem Does This Paper Address?

Modern CAVs rely on a suite of sensors (LiDAR, Radar, ultrasonic, cameras) to perform Cooperative Adaptive Cruise Control (CACC). However, these sensors are notoriously vulnerable. Attackers can use inexpensive hardware to "spoof" signals—for example, making a LiDAR believe an object is closer or further away than it actually is.

Existing security often focuses on cryptography (making sure the message isn't tampered with), but that doesn't help if the sensor itself is being fed fake physical data. The challenge is: how do you know which sensor to trust when they all provide different readings? And how do you ensure the car’s steering and braking (the control loop) don't overreact to the resulting errors?

Methods

The authors' approach relies on Sensor Redundancy. If you have $N$ sensors measuring the same variable (like the distance to the car in front), you have a baseline for truth.

  1. Reconstructability Analysis: They prove mathematically that as long as the number of attacked sensors ($q$) is less than half of the total sensors ($N/2$), the true value can be recovered.
  2. The Fusion Algorithm: Instead of a simple average (which an attacker could easily skew by providing a massive fake value), the algorithm looks at all possible subsets of sensors. It calculates the average of each subset and identifies the one where the sensors are most "in agreement" (minimum maximum deviation).
  3. Attack Detection and Isolation: Once a "trusted" estimate is found, the system compares individual sensor readings against it. If a sensor deviates beyond a calculated threshold (based on known noise bounds), it is flagged and isolated.
  4. Robust $H_\infty$ Control: The authors designed a controller specifically for vehicle platoons. It treats the remaining estimation error and communication noise as disturbances, using $H_\infty$ theory to ensure "string stability"—meaning a small error at the front of the car line doesn't amplify into a massive pile-up at the back.

Datasets

Not discussed in the paper. The authors do not use real-world driving datasets. Instead, they evaluate their framework using numerical simulations and synthetic data generated to model vehicle dynamics and sensor noise (Gaussian and Uniform distributions).

Key Insights and Ideas

  • The "Less Than Half" Rule: The most critical insight is the formal proof that $q < N/2$ is the breaking point. If an attacker controls 51% of your distance sensors, the system can no longer distinguish the "lie" from the "truth."
  • Bounded Error Independence: A major "aha moment" in the math (Theorem 2) shows that the fusion error is bounded by a constant factor (3x) of the maximum sensor noise. Crucially, this error is independent of the attack signal. This means no matter how wildly an attacker spoofs a sensor, they cannot pull the fused estimate beyond a small, safe margin.
  • Time-Varying Resilience: Unlike many previous models that assume the same sensors are always the ones under attack, this framework handles attackers who switch targets mid-drive to save energy or bypass simple filters.

Key Results

  • Theoretical Guarantee: The paper proves that the fused measurement error $e(t)$ satisfies $|e(t)| \leq 3||\nu||_\infty$, where $\nu$ is the measurement noise.
  • 100% Detection: In a simulation with three sensors (LiDAR, Radar, Ultrasonic), the isolation algorithm achieved a 100% success rate in detecting which sensor was being spoofed.
  • Platoon Stability: In a 5-vehicle platoon simulation, the $H_\infty$ controller successfully maintained the desired following distance even when sensors were being randomly attacked across different vehicles. The "string behavior" remained stable, preventing the "accordion effect" often seen in traffic.

Limitations and Future Work

  • The Redundancy Requirement: The framework requires $N \geq 3$ sensors for the same variable to survive even one attack. For many budget-conscious automotive designs, having three of every sensor might be cost-prohibitive.
  • Computational Complexity: The algorithm involves checking subsets of sensors. While manageable for a few sensors, the combinatorial complexity could grow if a system uses a very high number of redundant inputs.
  • Noise Bounds: The detection and isolation phase assumes the bounds of the "normal" sensor noise are known. In highly unpredictable weather (heavy rain/snow), these bounds might shift, potentially leading to false positives where a "noisy" sensor is mistaken for an "attacked" one.
  • Future Direction: The authors suggest extending this to more complex, non-linear vehicle models and exploring how to handle cases where more than half of the sensors might be compromised.

Reference: Yang, T., & Lv, C. (2021). A Secure Sensor Fusion Framework for Connected and Automated Vehicles under Sensor Attacks. arXiv:2103.00883v1.

Summary updated

Abstract

As typical applications of cyber–physical systems (CPSs), connected and automated vehicles (CAVs) are able to measure the surroundings and share local information with the other vehicles by using multimodal sensors and wireless networks. CAVs are expected to increase safety, efficiency, and capacity of our transportation systems. However, the increasing usage of sensors has also increased the vulnerability of CAVs to sensor faults and adversarial attacks. Anomalous sensor values resulting from malicious cyberattacks or faulty sensors may cause severe consequences or even fatalities. In this article, we increase the resilience of CAVs to faults and attacks by using multiple sensors for measuring the same physical variable to create redundancy. We exploit this redundancy and propose a sensor fusion algorithm for providing a robust estimate of the correct sensor information with bounded errors independent of the attack signals, and for attack detection and isolation. The proposed sensor fusion framework is applicable to a large class of security-critical CPSs. To minimize the performance degradation resulting from the usage of the estimation for control, we provide an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$H_{\infty }$ </tex-math></inline-formula> controller for cooperative adaptive cruise control-equipped CAVs. The designed controller is capable of stabilizing the closed-loop dynamics of each vehicle in the platoon while reducing the joint effect of estimation errors and communication channel noise on the tracking performance and string behavior of the vehicle platoon. Numerical examples are presented to illustrate the effectiveness of our methods.
BibTeX
@article{2022secure,
  author = {Chen Lv (吕辰) and Tianci Yang and Chen Lv},
  title = {{A Secure Sensor Fusion Framework for Connected and Automated Vehicles Under Sensor Attacks}},
  journal = {IEEE Internet of Things Journal},
  year = {2022},
  doi = {10.1109/jiot.2021.3101502},
}

Loading related content...

Figure