Observability-based consistent EKF estimators for multi-robot cooperative localization

In this paper, we investigate the consistency of extended Kalman filter (EKF)-based cooperative localization (CL) from the perspective of observability. We analytically show that the error-state system model employed in the standard EKF-based CL always has an observable subspace of higher dimension than that of the actual nonlinear CL system. This results in unjustified reduction of the EKF covariance estimates in directions of the state space where no information is available, and thus leads to inconsistency. To address this problem, we adopt an observability-based methodology for designing consistent estimators in which the linearization points are selected to ensure a linearized system model with observable subspace of correct dimension. In particular, we propose two novel observabilityconstrained (OC)-EKF estimators that are instances of this paradigm. In the first, termed OC-EKF 1.0, the filter Jacobians are calculated using the prior state estimates as the linearization points. In the second, termed OC-EKF 2.0, the linearization points are selected so as to minimize their expected errors (i.e., the difference between the linearization point and the true state) under the observability constraints. The proposed OC-EKFs have been tested in simulation and G.P. Huang ( ) · N. Trawny · S.I. Roumeliotis Department of Computer Science and Engineering, University of Minnesota, Minneapolis, MN 55455, USA e-mail: [email protected] N. Trawny e-mail: [email protected] S.I. Roumeliotis e-mail: [email protected] A.I. Mourikis Department of Electrical Engineering, University of California, Riverside, CA 92521, USA e-mail: [email protected] experimentally, and have been shown to significantly outperform the standard EKF in terms of both accuracy and consistency.