Cyber-Resilient Multi-Robot Systems: Toward Secure Autonomous Networks

<p dir="ltr">The reliability and safety of multi-robot systems (MRSs) are increasingly critical as such systems are deployed in real-world missions such as autonomous surveillance, infrastructure inspection, and disaster response. These systems rely on continuous communication and local sensing to achieve distributed coordination and consensus. However, this same interconnectivity exposes MRSs to vulnerabilities such as sensor faults, communication failures, and deliberate cyberattacks, which can corrupt shared information and compromise the collective objective of the system. Ensuring resilience in these settings requires not only the ability to detect and identify compromised agents, but also to guarantee stable network behavior even when adversarial or faulty information is introduced. Achieving this in a decentralized manner—without dependence on a central coordinator or trusted nodes—remains a central challenge in multi-agent robotics and distributed optimization.</p><p dir="ltr">The first part of this thesis addresses this challenge by introducing an anchor-free, range-based integrity monitoring framework for MRSs that detects, identifies, and reconstructs the effects of cyberattacks and system faults using only information inter-robot range measurements. By formulating the problem as a distributed optimization solved via Sequential Convex Programming (SCP) and the Alternating Direction Method of Multipliers (ADMM), the proposed approach enables real-time, scalable, and fully distributed fault detection without relying on any pre-defined anchors or trusted agents. A threshold-based cold-start mechanism is introduced to ensure robustness against sensor noise and changes in network topology, maintaining algorithmic stability under realistic, time-varying conditions. The framework is validated through extensive numerical simulations and mixed-reality experiments involving heterogeneous UAV swarms subjected to GNSS spoofing attacks, demonstrating effective detection, identification, and localization of compromised agents.</p><p dir="ltr">While this framework effectively detects and reconstructs faults, subsequent analysis revealed a critical vulnerability: falsified or inconsistent network information can bias the distributed optimization process itself, leading to divergence or erroneous consensus among nominal agents. To overcome this limitation, the second part of the thesis develops the Dynamically Weighted ADMM (DW-ADMM) algorithm, a Byzantine-resilient extension of ADMM that dynamically reweights communication edges to suppress the influence of malicious nodes. Theoretical results establish convergence to the global optimum in error-free conditions and bounded performance under Byzantine threats. Simulation studies confirm that DW-ADMM maintains consensus and optimization accuracy even when a subset of agents transmits adversarial data.</p><p dir="ltr">Together, these contributions form a comprehensive framework for resilient distributed estimation and optimization in multi-robot systems, bridging fault detection, attack identification, and Byzantine-tolerant coordination under a unified theoretical and experimental foundation.</p>