Distributed Robust Consensus Using RANSAC and Dynamic Opinions

Sensor networks must be able to fuse nodes' perceptions in a reliable way to reach a trustworthy consensus. Data association mistakes and measurement outliers are some of the factors that can contribute to incorrect perceptions and considerably affect consensus values. In this paper, we present a novel distributed scheme for robust consensus in autonomous sensor networks. The proposed method builds on random sampling consensus to exploit measurement redundancy, and enables the network to determine outlier observations with local communications. To do this, different hypotheses are generated and voted for using distributed averaging. In our approach, nodes can change their opinion as the hypotheses are computed, making the voting process dynamic. Assuming that enough hypotheses are generated to have at least one composed exclusively by inliers, we show that the method converges to the maximum likelihood of all the inlier observations under some natural conditions. We present several simulations and examples with real information that demonstrate the good performance of the proposed algorithm.