Hybrid Learning-Based Resilient Formation Control for Multi-Vehicle Systems Under Distributed Denial-of-Service Attacks

This paper investigates secure formation control for unknown networked multi-vehicle systems under denial-of-service (DoS) attacks. Unlike most existing studies that assume a unified attack model across all inter-vehicle communication channels, we propose an asynchronous, distributed DoS attack strategy targeting individual communication links. Specifically, we develop a resilient distributed observer capable of withstanding multi-channel asynchronous DoS attacks. This observer simultaneously provides both specific secure state estimation and output tracking references for each vehicle by introducing the concept of channel-dependent decay rates. Building upon the estimated information, we introduce a novel hybrid policy learning algorithm that combines off-policy and on-policy learning mechanisms. This hybrid approach enables data-driven derivation of decentralized formation control policies while overcoming key limitations of traditional methods, including the requirement for initial stabilizing policies and limitations in dynamic optimization capabilities. Finally, numerical simulations of networked multi-vehicle systems demonstrate the effectiveness of our proposed methodology.