Fault-Tolerant Covariance Intersection for Localizing Robot Swarms

Abstract

This paper examines the important problem of cooperative localization in robot swarms, in the presence of unmodeled errors experienced by real sensors in hardware platforms. Many existing methods for cooperative swarm localization rely on approximate distance metric heuristics based on properties of the communication graph. We present a new cooperative localization method that is based on a rigorous and scalable treatment of estimation errors generated by peer-to-peer sharing of relative robot pose information. Our approach blends Covariance Intersection and Covariance Union techniques from distributed sensor fusion theory in a novel way, in order to maintain statistical estimation consistency for cooperative localization errors. Experimental validation results show that this approach provides both reliable and accurate state estimation results for Droplet swarms in scenarios where other existing swarm localization methods cannot.