A Resilient Mean-Field Game Framework for Distributed Approximate Nash Equilibrium in Multi-Vehicle Systems under Attack

The practical deployment of multi-vehicle coordination systems faces dual challenges: the exponential computational complexity of N-player game theory and the critical vulnerability to cyber-attacks on communication networks. Existing approaches often address these issues in isolation, lacking a unified solution that is both scalable and secure. This paper introduces the Resilient Mean-Field Game (R-MFG) framework, a novel distributed control architecture designed to holistically overcome these barriers. R-MFG makes three key contributions: (1) it leverages Mean-Field Game (MFG) theory to ensure computational complexity scales with local traffic density, not the total number of vehicles; (2) it integrates an Extended State Observer (ESO) to provide model-agnostic resilience against False Data Injection (FDI) attacks by treating them as lumped disturbances; and (3) it employs a dual-layer optimization for real-time decision-making. We validate the R-MFG framework in a high-fidelity, data-driven simulation using the nuScenes dataset. The results demonstrate that R-MFG maintains safe and stable trajectories under attack, whereas a non-resilient baseline's behavior is compromised. Furthermore, we empirically confirm the framework's linear scalability with local density. The R-MFG framework represents a significant step towards enabling safe, scalable, and secure autonomous transportation systems.