Multi-agent adaptive filtering with outlier mitigation using constrained mixture distributions

Abstract

To enhance multi-agent state estimation under measurement noise with unknown (potentially time-varying) covariance and polluted by outliers, we employ a Bernoulli–Gaussian model of measurement noise with constrained inverse-Wishart distributions for the unknown covariances. Building upon this model, we propose a novel robust adaptive constrained filter as well as a distributed multi-sensor extension integrating variational Bayesian and hybrid consensus approaches. Simulation results in a target tracking scenario demonstrate the effectiveness of the proposed filter in addressing state estimation challenges arising from unknown measurement noise statistics and the presence of outliers.