Collaborative localization of robotic wheeled walkers using interlaced Extended Kalman Filters

Abstract

This paper deals with the problem of localizing a group of robots. Each robot is equipped with an autonomous self-localization system based on both high-rate odometers and a front camera detecting sporadically simple visual landmarks placed on the floor at known positions and with a known attitude within a given 2D reference frame. In addition, the robots can share their own states to improve localization accuracy. In this paper, after reviewing the general problem of collaborative localization, two potential strategies based on a distributed Interlaced Extended Kalman Filter (IEKF) are compared. In the first one, localization is refined through high-rate mutual measurements of Euclidean distance between pairs of robots. The distance values are obtained using an omni-directional wireless ranging system of limited accuracy. In the second one, an on-board Kinect-like front RGB-D vision system is assumed to measure the agent's relative position with respect to other robots in view at a lower rate, but with higher accuracy. To the best of authors' knowledge, this performance comparison is new, since most of previous studies on collaborative localization are overoptimistic as implicitly assume that mutual agent measurements are not intermittent and are available at all sampling times.