Distributed resilient consensus: a non-parametric approach

In this paper, two parameter-independent fault-tolerant consensus algorithms are proposed to address the consensus problem in the presence of misbehaving agents. The first algorithm relies on adaptively estimating the number of faulty agents in the network by using a distributed fault-detection scheme. It is shown that this algorithm converges if the network of non-faulty agents is (f+1)-robust, where f is the number of faulty agents in the network. The second algorithm is a non-parametric Mean-Subsequence-Reduced algorithm whose convergence is guaranteed if the network of non-faulty nodes is (f+1)-robust and all non-faulty nodes have the same number of in-neighbours. Neither algorithm requires initial knowledge on the number of faulty agents in the network. The efficacy of the algorithms are illustrated with simulation results.