Criticality-Based Dynamic Topology Optimization for Enhancing Aerial-Marine Swarm Resilience

Ruiyang Huang, Haocheng Wang, Yixuan Shen, Ning Gao, Qiang Ni, Shi Jin, Yifan WuarXiv 2025 · arXiv · OpenReview

TL;DR

This paper introduces a novel framework, SurBi-Ranking, that uses graph convolutional networks to identify critical components in aerial-marine swarms. This criticality information is then used with a multi-objective optimization algorithm (NSGA-III) to dynamically reconfigure swarm topologies, significantly improving resilience against targeted attacks and ensuring mission success.

What Problem Does This Paper Address?

Heterogeneous swarms of aerial vehicles (UAVs) and marine surface vehicles (USVs) are increasingly used for critical applications like search and rescue. However, these swarms are vulnerable in adversarial environments where targeted attacks (e.g., jamming, physical disruption) can cripple communication and lead to cascading failures. Existing resilience approaches often rely on static or single-layer models, failing to capture the complex interdependencies between the swarm's physical structure, communication links, and mission tasks. This paper addresses the challenge of dynamically optimizing swarm topology to maintain operational effectiveness and resilience in the face of sophisticated, multi-layer attacks.

Methods

The authors propose a two-step framework:

  1. Three-Layer Network Model & Adversarial Model: They first model the swarm as a three-layer network:
  2. Structural Layer: Represents physical connections and formations between vehicles.
  3. Communication Layer: Models wireless links, considering factors like distance, latency, and bit error rate.
  4. Task Layer: Captures mission dependencies and requirements. This is complemented by a sophisticated adversarial model that simulates an attacker with situational awareness capable of targeted node and edge disruptions across these layers.

  5. SurBi-Ranking for Criticality Assessment: This is the core novelty. They introduce the SurBi-Ranking method, which combines:

  6. Graph Convolutional Networks (GCNs): To learn node representations from topological features.
  7. Birnbaum Importance: A measure from reliability engineering adapted to assess how removing a node impacts overall network connectivity.
  8. Surrounding Node Influence: Considers the importance of a node's neighbors, capturing local network context. By fusing these, SurBi-Ranking dynamically evaluates and ranks the criticality of nodes and edges in real-time.

  9. Multi-Objective Topology Optimization: The NSGA-III algorithm is employed to optimize the swarm's topology. The objectives are to balance:

  10. Global Connectivity (f1)
    • Global Communication Success Rate (f2)
    • Global Vulnerability (f3)
    • Reconfiguration Cost (f4) (introduced for dynamic adjustment)

Datasets

The paper uses three datasets for evaluation:

  1. Planar Layer Network (PLN): A scale-free network with 1000 nodes, used for initial testing of SurBi-Ranking and topology optimization under attack.
  2. Multi-Phase Mission Dataset: Similar to PLN but models a five-phase mission, where network topology dynamically reconfigures between phases to meet specific requirements. This dataset is crucial for evaluating the multi-phase resilience.
  3. 3D Contested Environment Dataset: A smaller dataset (50 nodes: 30 UAVs, 20 USVs) simulating a 3D environment with realistic constraints (e.g., altitude, inter-node distances) and multi-phase missions, designed for more realistic topological optimization.

Key Insights and Ideas

  • Cross-Layer Dependencies are Crucial: The paper strongly emphasizes that understanding and modeling the interplay between structural, communication, and task layers is paramount for effective resilience, a gap often overlooked in prior work.
  • Dynamic Criticality is Key: Static centrality measures are insufficient. SurBi-Ranking's ability to dynamically assess node and edge importance in real-time, considering both intrinsic and neighborhood influence, is a significant advantage.
  • Targeted Attacks Exploit Criticality: The research demonstrates that deliberate attacks targeting components identified by SurBi-Ranking lead to much faster network degradation than random attacks or attacks based on less sophisticated metrics.
  • Topology Optimization Adapts to Threats: The NSGA-III algorithm, guided by SurBi-Ranking, can find optimal topologies that balance multiple objectives, including resilience against specific attack scenarios and minimizing reconfiguration overhead.
  • Phase-Specific vs. Global Criticality: The analysis highlights that some nodes are critical only within a specific mission phase, while others are globally critical across all phases, necessitating a nuanced approach to defense.

Key Results

  • SurBi-Ranking Superiority: Compared to K-Shell and Katz centrality, SurBi-Ranking identifies critical nodes and edges more accurately, leading to significantly steeper connectivity degradation when these components are attacked.
  • Reduced Connectivity Degradation: The proposed optimization approach, prioritizing SurBi-Ranked critical components under attack, reduces natural connectivity degradation by approximately 30%.
  • Improved Mission Success: The optimized topologies achieve higher mission success rates.
  • Lower Reconfiguration Costs: The framework balances resilience with communication reconfiguration costs.
  • Optimal Parameter 'r': Experiments showed that an optimal value for the weighting parameter 'r' in SurBi-Ranking (around 0.3 in their tests) is crucial for accurate criticality assessment and is topology-dependent.

Limitations and Future Work

  • Parameter 'r' Calibration: The optimal weighting parameter 'r' for SurBi-Ranking is topology-dependent and requires case-specific calibration, which could be a practical challenge.
  • Computational Complexity: While not explicitly detailed as a limitation, dynamic topology optimization and GCN training can be computationally intensive, especially for very large swarms.
  • Future Work: The authors suggest further research into more sophisticated adversarial models, exploring adaptive learning for 'r' calibration, and extending the framework to more complex swarm behaviors and environments.

Summary updated

Abstract

Heterogeneous marine-aerial swarm networks encounter substantial difficulties due to targeted communication disruptions and structural weaknesses in adversarial environments. This paper proposes a two-step framework to strengthen the network's resilience. Specifically, our framework combines the node prioritization based on criticality with multi-objective topology optimization. First, we design a three-layer architecture to represent structural, communication, and task dependencies of the swarm networks. Then, we introduce the SurBi-Ranking method, which utilizes graph convolutional networks, to dynamically evaluate and rank the criticality of nodes and edges in real time. Next, we apply the NSGA-III algorithm to optimize the network topology, aiming to balance communication efficiency, global connectivity, and mission success rate. Experiments demonstrate that compared to traditional methods like K-Shell, our SurBi-Ranking method identifies critical nodes and edges with greater accuracy, as deliberate attacks on these components cause more significant connectivity degradation. Furthermore, our optimization approach, when prioritizing SurBi-Ranked critical components under attack, reduces the natural connectivity degradation by around 30%, achieves higher mission success rates, and incurs lower communication reconfiguration costs, ensuring sustained connectivity and mission effectiveness across multi-phase operations.
BibTeX
@misc{huang2025criticality,
  author = {Ruiyang Huang and Haocheng Wang and Yixuan Shen and Ning Gao and Qiang Ni and Shi Jin and Yifan Wu},
  title = {{Criticality-Based Dynamic Topology Optimization for Enhancing Aerial-Marine Swarm Resilience}},
  year = {2025},
  eprint = {2508.00688},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.00688},
}
Figure