Ripple effect of cooperative attacks in multi-agent systems: Results on minimum attack targets

Abstract

From the security of multi-agent systems (MASs), this paper investigates false-data injection attacks that cooperatively compromise communication links to stealthily destabilize the leader-following control process. Different from centralized systems, the communication coupling of MASs enables attackers to have more complex adverse influences. In view of this, we focus on how attacks perform ripple effect by spreading their influences to the agents receiving no compromised communication data. First of all, the existence of cooperative false-data injection (CFDI) attacks is studied based on the spectral radius of system matrices. Further, we explore attack ripple effect from the relations between impaired agents and attack targets against communication links. Therefore, with the geometric and algebraic properties of edge-to-node incidence matrices, a necessary and sufficient condition of minimum attack targets is developed. On this basis, the protection targets of communication data are qualitatively analyzed to resist CFDI attacks in MASs. The numerical simulations conducted on a group of two-link planar robots illustrate the theoretical results.