Trust is the Cure to Distributed Consensus with Adversaries

Extensive research efforts have been devoted to distributed consensus with adversaries. Many diverse applications drive this increased interest in this area including distributed collaborative sensor networks, sensor fusion and distributed collaborative control. We consider the problem of detecting Byzantine adversaries in a network of agents with the goal of reaching consensus. We propose a novel trust model that establishes both local trust based on local evidences and global trust based on local exchange of local trust values. We describe a trust-aware consensus algorithm that integrates the trust evaluation mechanism into the traditional consensus algorithm and propose various local decision rules based on local evidence. To further enhance the robustness of trust evaluation itself, we also provide a trust propagation scheme in order to take into account evidences of other nodes in the network. The algorithm is flexible and extensible to incorporate more complicated designs of decision rules and trust models. To demonstrate the power of our trust propagation scheme, we provide theoretical security performance in terms of miss detection rate and false alarm rate under regular trust graph and relaxed security performance bound under general trust graph. In addition, we demonstrate through simulation that the trust-aware consensus algorithm can effectively detect Byzantine adversaries and excluding them from consensus iterations even in sparse networks with connectivity less than 2f + 1, where f is the number of adversaries. These results can be applied for fusion of trust evidences as well as for sensor fusion when malicious sensors are present like for example in power grid sensing and monitoring.