Encryption–decryption-based distributed fault-tolerant consensus tracking control for multi-agent systems

Yiliu Liu, Chun Liu, Xiaofan Wang, Jianglin Lan, Xiaoqiang RenISA 2024 · DOI

This paper, recently published in ISA Transactions, presents a robust framework for managing multi-agent systems (MASs) that are simultaneously plagued by hardware failures and security vulnerabilities.

TL;DR

The authors propose a distributed control strategy that achieves "dual guarantees": it compensates for physical actuator faults (like bias or drift) while protecting agent privacy from external eavesdroppers. By combining decentralized observers with a dynamic encryption-decryption scheme, the system ensures accurate leader-tracking without leaking sensitive state information over insecure communication channels.

What Problem Does This Paper Address?

Multi-agent systems—such as swarms of drones or fleets of autonomous vehicles—rely on constant communication to stay synchronized. This creates two major points of failure:

  1. Physical Faults: Actuators can suffer from "bias" (a constant offset) or "drift" (a time-varying error), leading to collisions or mission failure.
  2. Cyber Vulnerabilities: Passive eavesdroppers can listen in on communication links to steal the physical states of the agents.

While previous research has tackled these issues individually, doing both at once is difficult. Privacy-preserving methods like "noise injection" often degrade control accuracy, and many fault-tolerant controllers assume communication links are perfectly secure. This paper seeks a way to maintain high-precision tracking (consensus) in discrete-time systems while keeping data confidential and communication efficient.

Methods

The researchers developed a three-part architecture to solve this:

  • State and Fault Augmentation: The authors treat actuator faults as additional "hidden" state variables. They designed decentralized observers for each agent that use only local input/output data to estimate both the agent's true state and the magnitude of any active faults.
  • Encoding-Decoding EDS (Encryption-Decryption Strategy): To thwart eavesdroppers, agents do not broadcast their raw state estimates. Instead, they transmit a "cipher" based on a dynamic symmetric key $g(t)$ that decays over time. This strategy uses a quantizer to send only the prediction error between the current state and a predicted value, which naturally compresses the data.
  • Active Fault-Tolerant Consensus Tracking Control (FCTC): The controller uses the decrypted information from neighbors and the local fault estimates to "offset" the impact of the faults in real-time.

The stability of this complex interaction was proven using Modified Discrete-time Algebraic Riccati Equations (MDARE), providing clear criteria for choosing controller gains and encryption parameters.

Datasets

The paper does not use external real-world datasets. Instead, the authors validate their approach using a numerical simulation of a networked fleet consisting of one leader and four follower vehicles. The vehicles are modeled as double integrators (representing position and velocity dynamics), a standard benchmark in control theory.

Key Insights and Ideas

  • Predictive Privacy: A standout feature of their EDS is that the encryption variable acts as a predictor. By quantizing the difference between the actual state and the prediction, the system transmits much smaller values. This simultaneously hides the true state from hackers and reduces the "communication load" on the network.
  • Active vs. Passive Compensation: Unlike "passive" controllers that are simply designed to be robust, this "active" controller uses real-time fault estimates to aggressively cancel out errors as they happen.
  • Decoupled Design: The authors used "compatible norms" to prove that the controller, the observer, and the encryption key can be designed somewhat independently while still guaranteeing that the whole system will converge to a steady state (consensus).

Key Results

  • Fault Recovery: In simulations, when agents were hit with bias faults (at 15s and 40s) and drift faults (at 30s and 50s), the decentralized observers identified the faults almost instantly.
  • Tracking Accuracy: Despite the "noise" introduced by quantization and encryption, the state tracking error for all followers converged to zero, achieving perfect consensus with the leader.
  • Efficiency: The "Original Data" (raw states) in the simulation tended to grow indefinitely due to the system's unstable nature, but the "Encrypted Data" remained bounded and small. This confirms the strategy alleviates communication congestion while maintaining security.

Limitations and Future Work

The authors acknowledge a few areas for further exploration:

  • Linearity: The current model assumes linear discrete-time dynamics. Real-world robots often have highly nonlinear behavior (e.g., aerodynamics) that would require more complex observers.
  • Internal Threats: The paper focuses on external eavesdroppers. It does not account for "Byzantine" agents—malicious insiders that are part of the swarm and broadcast false data to disrupt the consensus.
  • Heterogeneity: The study assumes all agents are identical. Future work could look at heterogeneous MASs where different types of robots (e.g., aerial and ground) must work together under the same security constraints.

Summary updated

BibTeX
@inproceedings{liu2024encryptiondecryption,
  author = {Yiliu Liu and Chun Liu and Xiaofan Wang and Jianglin Lan and Xiaoqiang Ren},
  title = {{Encryption–decryption-based distributed fault-tolerant consensus tracking control for multi-agent systems}},
  booktitle = {ISA},
  year = {2024},
  doi = {10.1016/j.isatra.2024.08.015},
}
Figure