A Multigrid Approach for Accelerating Relaxation-Based SLAM

Abstract

This paper addresses the problem of simultaneouslocalisation and mapping (SLAM) by a mobile robot. Anincremental SLAM algorithm is introduced that is derivedfrom socalled multigrid methods used for solving partialdifferential equations. The approach overcomes therelatively slow convergence of previous relaxation methodsbecause it optimizes the map at multiple levels ofresolution. The resulting algorithm has an update time thatis linear in the number of mapped features, even whenclosing very large loops, and offers advantages in handlingnon-linearities compared to previous approaches.Experimental comparisons with alternative algorithms usingtwo well-known data sets are also presented. 1