Active Resilience Control for UAV Swarms
Official implementation of the paper: Active resilience control for UAV swarms: A closed-loop framework integrating collaborative perception and dynamic metrics Reliability Engineering & System Safety (RESS), 2026
Paper
- Paper DOI: https://doi.org/10.1016/j.ress.2026.112400
- Model DOI: https://doi.org/10.57967/hf/8105
Code
GitHub Repository: https://github.com/zyfgators/ARCS
Abstract
Task resilience has become a critical metric for assessing the survivability and mission assurance of unmanned aerial vehicle (UAV) swarms in complex dynamic environments. However, existing research lacks dynamic resilience measurement capabilities and suffers from a disconnection between assessment metrics and control actions. To address these gaps, a theoretical framework linking interference perception, dynamic metrics, and closed-loop control is established. First, a Spatio-Temporal Collaborative Localization Neural Network (STCL-NN) is designed to localize unknown interference sources, introducing a multi-dimensional confidence index to quantify estimation credibility. Subsequently, an interference damage dynamics model is developed to transform physical interference into an equivalent reduction in task payload capability. Combined with a sliding-window prediction mechanism, a dynamic resilience measurement system is constructed to enable real-time evaluation and forward-looking prediction. Finally, an Active Resilience Optimal Control strategy is designed utilizing Pontryagin’s Minimum Principle (PMP) to implicitly guarantee that the system’s operational performance satisfies the global resilience safety baseline. Simulations demonstrate the method’s superiority: (1) Perception Credibility: The confidence index mechanism effectively filters estimation uncertainty, providing a trustworthy data foundation for resilience measurement; (2) Resilience Assurance: The system maintains a 100% satisfaction rate regarding the mission-level resilience requirements; (3) Operational Efficiency: Compared to conventional control strategies, the proposed strategy reduces the actuator triggering frequency by approximately 64%, extending hardware longevity, while incurring a marginal time cost under 1.5% relative to the theoretical lower bound. This marks a critical shift in the resilience preservation of cluster systems, from passive robustness to active regulation.