Unifying Consensus and Covariance Intersection for Efficient Distributed State Estimation over Unreliable Networks

This thesis studies the problem of recursive distributed state estimation over unreliable networks. The main contribution is to fuse the independent and dependent information separately. Local estimators communicate directly only with their immediate neighbors and nothing is assumed about the structure of the communication network, specifically it need not be connected at all times. The proposed estimator is a Hybrid one that fuses independent and dependent (or correlated) information using a distributed averaging and iterative conservative fusion rule respectively. It will be discussed how the hybrid method can improve estimators's performance and make it robust to network failures. The content of the thesis is divided in two main parts. In the first part I study how this idea is applied to the case of dynamical systems with continuous state and Gaussian noise. I establish bounds for estimation performance and show that my method produces unbiased conservative estimates that are better than Iterative Covariance Intersection (ICI). I will test the proposed algorithm on an atmospheric dispersion problem, a random linear system estimation and finally a target tracking problem. In the second part, I will discuss how the hybrid method can be applied to distributed estimation on a Hidden Markov Model. I will discuss the notion of conservativeness for general probability distributions and use the appropriate cost function to achieve improvement similar to the first part. The performance of the proposed method is evaluated in a multi-agent tracking problem and a high dimensional HMM and it is shown that its performance surpasses the competing algorithms