Control of Multilayer Mobile Autonomous Systems in Adversarial Environments: A Games-in-Games Approach

Juntao Chen, Quanyan Zhu

TL;DR

This paper introduces a "games-in-games" framework to design secure, multi-layer mobile autonomous systems (MAS) that can maintain network connectivity even when attacked. It proposes a novel "meta-equilibrium" solution concept and an iterative algorithm to achieve this, demonstrating its effectiveness in simulations against jamming and GPS spoofing attacks.

What Problem Does This Paper Address?

Modern applications increasingly rely on multiple, interconnected mobile autonomous systems (MAS), forming "multi-layer" networks. Think of drones and ground robots working together in disaster response or battlefield scenarios. The crucial challenge is maintaining reliable communication and coordination (network connectivity) within and between these layers, especially when malicious actors (attackers) try to disrupt these networks. Traditional control methods often assume a single, cooperative network or a fully centralized system, which doesn't scale to the complexity and adversarial nature of these multi-layer MAS. This paper tackles the problem of designing decentralized control strategies for these multi-layer MAS that are robust to cyberattacks, specifically jamming and GPS spoofing, while accounting for the uncoordinated decision-making of different network operators.

Methods

The core of this work is the games-in-games framework. This framework models the interactions between:

  1. Two network operators (players): Each operator controls their own layer of MAS (e.g., drones or ground robots) and aims to maximize the overall network connectivity. They operate independently, leading to a Nash game between them.
  2. A jamming attacker: This adversary tries to minimize the network connectivity by removing critical communication links. The network operators anticipate this and design their strategies accordingly, leading to a Stackelberg game where the operator is the leader and the attacker is the follower.

To capture this complex interplay, the authors propose a meta-equilibrium solution concept. This concept ensures that:

  • The attacker's strategy is a best response to the operators' actions (Stackelberg equilibrium).
  • The operators' strategies form a Nash equilibrium, meaning neither operator can improve the overall connectivity by unilaterally changing their strategy.

For practical implementation, they develop a resilient and decentralized iterative algorithm. This algorithm allows network operators to update robot positions iteratively, aiming to maximize algebraic connectivity while considering worst-case attack scenarios. The algorithm is designed to converge to the meta-equilibrium.

Datasets

Not discussed in the paper. The evaluation relies on simulated scenarios with defined parameters for robot movement, communication strength, and attack models.

Key Insights and Ideas

  • Games-in-Games for MAS Security: The paper successfully frames the complex problem of multi-layer MAS control under adversarial conditions as a combination of Nash and Stackelberg games, providing a holistic approach.
  • Meta-Equilibrium: This novel solution concept is crucial for characterizing the system's behavior, balancing decentralized control, adversarial anticipation, and overall network performance.
  • Agile Resilience: The proposed algorithm not only maintains connectivity during attacks but also demonstrates the ability to recover quickly once the attack is removed, highlighting the system's resilience.
  • Anticipation is Key: The work strongly suggests that proactively designing for potential attacks (rather than reacting after they occur) significantly improves network robustness and performance.
  • Non-Uniqueness of Equilibria: The authors show that multiple meta-equilibrium configurations can exist, implying that selecting the "best" equilibrium might be a future research direction.

Key Results

  • The proposed iterative algorithm converges asymptotically to a meta-equilibrium, meaning the MAS network reaches a stable, secure configuration.
  • Case studies demonstrate that the secure MAS network, designed with the games-in-games approach, remains connected even under jamming attacks that would disconnect a network designed without considering adversaries.
  • Simulations show that the system exhibits agile resilience to GPS spoofing attacks, recovering connectivity quickly after the attack is detected and removed.
  • The framework is shown to be effective for two-layer MAS networks, with potential for extension to more complex multi-layer scenarios.

Limitations and Future Work

  • Non-Uniqueness of Meta-Equilibrium: The paper acknowledges that multiple meta-equilibria can exist. Future work could focus on developing mechanisms for selecting the most desirable equilibrium (e.g., one with higher connectivity).
  • Centralized Planner: To guide the system towards a specific equilibrium, the authors suggest the need for an additional "central planner" with perfect information, which could be a direction for future research.
  • Heterogeneous Security Requirements: The current model assumes uniform security considerations. Future work could explore scenarios with different security requirements across layers or agents.
  • Sophisticated Attackers: The attacker model is primarily reactive (best response). Extending the framework to handle more sophisticated attackers who might plan over multiple time horizons is a potential avenue.
  • Real-World Deployment: While simulations are promising, real-world testing and validation would be the next logical step.

Summary updated

Abstract

A mobile autonomous system (MAS) becomes pervasive especially in the vehicular and robotic networks. Multiple heterogeneous MAS networks can be integrated together as a multilayer MAS network to offer holistic services. The network connectivity of the multilayer MAS plays an important role in the information exchange between agents within and across different layers of the network. In this article, we establish a games-in-games framework to capture the uncoordinated nature of decision making under adversarial environment at different layers. Specifically, each network operator controls the mobile agents in his own subnetwork and designs a secure strategy to maximize the global network connectivity by considering the behavior of jamming attackers that aim to disconnect the network. The solution concept of metaequilibrium is proposed to characterize the system-of-systems behavior of the autonomous agents. For online implementation of the control, we design a resilient algorithm that improves the network algebraic connectivity iteratively. We show that the designed algorithm converges to a metaequilibrium asymptotically. Finally, we use case studies of a two-layer MAS network to corroborate the security and agile resilience of the network controlled by the proposed strategy.
BibTeX
@article{chen2020control,
  author = {Juntao Chen and Quanyan Zhu},
  title = {{Control of Multilayer Mobile Autonomous Systems in Adversarial Environments: A Games-in-Games Approach}},
  journal = {IEEE Trans. Control. Netw. Syst.},
  year = {2020},
  url = {https://openreview.net/forum?id=XLYCQ0uETd},
}
Figure