A System-Theoretic Privacy-Informed Framework in Multi-Agent Systems

The problem of privacy-preserving in multi-agent systems corresponds to prohibiting the disclosure of agents' initial states while ensuring desired performance such as distributed average consensus. In this paper, a system-theoretic dynamic privacy-informed framework is developed. The proposed privacy framework relies on an obfuscation phase where a dynamic mask is inserted on the agents' state trajectories and masked outputs are exchanged amongst agents, rendering the physical states of an agent indiscernible by the other agents or external curious attackers (eavesdropper adversaries). Application of the proposed dynamic privacy scheme to well-known problems in multi-agent systems including average consensus and social opinion dynamics are presented.