Tolerance-Based Disruption-Tolerant Consensus in Directed Networks

This article addresses the problem of resilient consensus for multi-agent networks. Resilience is used here to distinguish disruptive agents from compliant agents which follow a given control law. We present an algorithm enabling efficient and resilient network consensus based on an inversion of the social dynamics of the Deffuant model with emotions. This is achieved through the exploitation of a dynamic tolerance linked to extremism and clustering, whereby agents filter out extreme non-standard opinions driving them away from consensus. This method is not dependent on prior knowledge of either the network topology or the number of disruptive agents, making it suitable for real-world applications where this information is typically unavailable.