Secure Cooperative Localization for Connected Automated Vehicles Based on Consensus

Xin Xia, Runsheng Xu, Jiaqi MaIEEE Sensors Journal 2023 · DOI

Abstract

In this paper, we present secure cooperative localization for connected automated vehicles (CAVs) based on consensus estimation through leveraging shared but possibly attacked sensory information from multiple adjacent vehicles. First, the communication topology between the CAVs, node kinematic model, and node measurement model for each vehicle are introduced. Then, a consensus Kalman information filter (CKIF) is applied to fuse the shared information from connected vehicles. Since the sensory information might be attacked, an attack detection algorithm based on the general likelihood ratio test (GLRT) is adopted. A delay-prediction framework is proposed to maintain the accuracy and real-time performance of the detection algorithm. Next, a rule-based attack isolation method is used to defend the attack. Finally, the proposed secure cooperative localization algorithm is validated in extensive numerical simulation experiments. The results confirm that leveraging information from multiple vehicles in a cooperative manner leads to better accuracy and resilience for vehicle localization under attacks.

Introduction: The Vulnerability of Cooperative Localization

Reliable localization is the backbone of Connected Automated Vehicles (CAVs). While individual vehicles use Global Navigation Satellite Systems (GNSS) and Inertial Measurement Units (IMUs), these systems are prone to failure in "urban canyons" or, more dangerously, to "drift attacks." In such attacks, an adversary injects gradual errors into GNSS or radar data to steer a vehicle off course without triggering immediate alarms.

Physical model of a vehicle platoon under sensory attack.

Cooperative localization—where vehicles share data to improve collective accuracy—offers a potential solution. However, it introduces a new vulnerability: if one vehicle is compromised, it can "poison" the location estimates of the entire platoon. The challenge addressed in this research is to reap the benefits of shared data while ensuring the network remains resilient against malicious actors.

Methodology: A Multi-Layered Defense Framework

The proposed system utilizes a multi-layered defense strategy to maintain localization integrity. The architecture integrates attack detection, isolation, and a distributed filtering mechanism.

System architecture for secure cooperative localization, featuring attack detection and the Consensus Kalman Information Filter.

Consensus Kalman Information Filter (CKIF)

Instead of relying on a centralized server, each vehicle runs its own filter. Vehicles share "information vectors" and "information matrices" to reach a consensus on their states. This distributed approach avoids single points of failure and allows the group to mathematically identify and ignore outliers through a "democratic" voting process.

Attack Detection and Isolation

To protect the CKIF from poisoned data, the system employs several key components:

  1. GLRT-Based Attack Detection: A General Likelihood Ratio Test (GLRT) monitors "innovations" (the difference between predicted and actual measurements). Significant deviations from expected Gaussian noise signal an attack.
  2. Delay-Prediction Framework: Sophisticated attack detection often requires a window of historical data, which can introduce dangerous latency. To solve this, the authors use the vehicle's onboard IMU to predict the current state while the detector processes the delayed data window, ensuring real-time responsiveness.
  3. Rule-Based Isolation: Once an attack is identified on a specific sensor or neighboring vehicle, the CKIF dynamically adjusts its covariance matrix to effectively "mute" that data source until the threat subsides.

Key Results and Insights

The framework was validated using extensive numerical simulations within the OpenCDA ecosystem, modeling platoons of four to ten vehicles. Key findings include:

  • Accuracy Boost: In a 10-vehicle platoon with undirected communication (where vehicles talk to both those in front and behind), the proposed CKIF improved localization accuracy by approximately 35.4% to 36.6% compared to standard directed cooperative methods.
  • Attack Resilience: The system successfully detected drift attacks within 0.1 seconds. In specific simulations, the CKIF achieved a Root Mean Square Error (RMSE) of 0.378m, significantly outperforming the Multi-Sensor Multi-Vehicle (MSMV) method (0.468m) and standard non-cooperative Kalman Filters (0.509m).
  • The IMU as an Anchor of Trust: A critical insight is the reliance on internal IMU data, which is much harder to spoof remotely than external GNSS or V2X signals. This internal data acts as a "bridge" during the detection process.

Conclusion and Future Work

While the current model demonstrates high resilience in longitudinal movement, future work is needed to address complex lateral maneuvers like lane changes. Additionally, real-world environmental factors, such as signal occlusion in dense urban environments, present ongoing challenges for maintaining the consensus chain. Nevertheless, this research provides a robust foundation for secure, cooperative autonomy in the face of increasing cyber-physical threats.

Summary updated

Abstract

In this paper, we present secure cooperative localization for connected automated vehicles (CAVs) based on consensus estimation through leveraging shared but possibly attacked sensory information from multiple adjacent vehicles. First, the communication topology between the CAVs, node kinematic model, and node measurement model for each vehicle are introduced. Then, a consensus Kalman information filter (CKIF) is applied to fuse the shared information from connected vehicles. Since the sensory information might be attacked, an attack detection algorithm based on the general likelihood ratio test (GLRT) is adopted. A delay-prediction framework is proposed to maintain the accuracy and real-time performance of the detection algorithm. Next, a rule-based attack isolation method is used to defend the attack. Finally, the proposed secure cooperative localization algorithm is validated in extensive numerical simulation experiments. The results confirm that leveraging information from multiple vehicles in a cooperative manner leads to better accuracy and resilience for vehicle localization under attacks.
BibTeX
@article{xia2023secure,
  author = {Xin Xia and Runsheng Xu and Jiaqi Ma},
  title = {{Secure Cooperative Localization for Connected Automated Vehicles Based on Consensus}},
  journal = {IEEE Sensors Journal},
  year = {2023},
  doi = {10.1109/jsen.2023.3312610},
}
Figure