Trustworthy Multi-UAV Collaboration: A Self-Supervised Framework for Explainable and Adversarially Robust Decision-Making

Ensuring trustworthiness in multi-UAV collaboration is essential for deploying autonomous aerial systems in safety-critical applications such as search and rescue, environmental monitoring, and infrastructure inspection. However, UAV decision-making remains opaque and susceptible to perception inconsistencies, sensor noise, and network uncertainties, undermining reliability in real-world scenarios. To address these challenges, we propose a self-supervised framework for explainable and robust multi-UAV decision-making, enabling UAVs to generate interpretable confidence assessments, verify internal consistency, and dynamically adjust decision thresholds based on environmental conditions and mission dynamics. Mutual verification through external consistency validation ensures alignment in perception and decision logic while mitigating the effects of sensor noise and adversarial perturbations. Additionally, our dynamic network adaptation mechanism adjusts confidence propagation weights and seamlessly integrates new agents, preserving decision stability despite fleet variations. We formalize this framework through rigorous mathematical modeling, proving that confidence updates remain bounded and self-regulating, that multi-UAV consensus is consistently achievable, and that system-wide decision adaptation remains stable under operational uncertainties. By enhancing interpretability, adaptability, and robustness, this framework lays the foundation for advancing trustworthy autonomous multi-agent systems in complex real-world applications.