Conservative Data Exchange for Decentralized Cooperative Localization

Abstract

Cooperative localization by multiple robots, performing decentralized navigation, is a challenging task in an enclosed environment due to the lack of access to the external facilities. If there is a cyclic update, they suffer from inconsistent estimates with higher confidence, namely overconfidence. This paper proposes a conservative data exchange approach for the cooperative localization, in which a fractional exponent is applied to each robot’s estimate before passing the information to the other robots. This method preserves the amplitude of the original information so that they can avoid falling into a wrong estimated state by cyclic updates. We also show that, when the local estimates are assumed to follow normal distributions, the proposed method behaves similarly to the covariance intersection (CI). A simulation has been conducted in a two-dimensional space to evaluate the proposed method by comparing it with other approaches – the naive and the CI-based methods. The results show that the proposed conservative data exchange approach outperforms the other methods.