Session 8: Localization and Mapping

Abstract

This session consisted of five papers in the area of localization and mapping addressing challenges that stem from the nonlinear motion and measurement models involved and the need to reliably recognize locations described with laser or visual data.

The first paper, “A Robust Method of Localization and Mapping Using Only Range” by Djugash and Singh, deals with the problem of simultaneous localization and mapping using range measurements. In this case, using an Extended Kalman filter (EKF) for estimating the robots’ pose and landmarks’ positions in Cartesian space is suboptimal and often leads to divergence and inconsistent estimates. This is due to the nonlinear measurement model that invalidates the Gaussian approximation. To address this issue, the authors introduce a new representation of the state vector in polar coordinates. The resulting estimator, the Relative Over-Parameterized (ROP)-EKF, is shown to be robust to large initialization errors and incorrect data associations. Additionally, it is able to operate with intermittent range measurements over large-scale experiments.