A comparison of SLAM implementations for indoor mobile robots

Abstract

This paper compares the performances of four different Simultaneous Localization and Mapping (SLAM) implementations: Teambotica1,5, ICP 6D-SLAM2,5, SGI-J DP SLAM3, and ERSP vSLAM4,5. All the tested implementations used odometry for dead-reckoning as a base position measurement, and all but vSLAM use lasers as a main sensor (ERSP vSLAM uses a video camera). The comparison was taken using two different types of robots in three different environments: an office corridor, an office entry area, and an indoor robot driving course. The comparison uses a novel approach for map accuracy evaluation, one of the key factors in evaluating the performance of a SLAM implementation. The results show that even for similar setups, the outcomes can vary significantly for the different implementations, suggesting that the underlying implementations likely play a significant role in the SLAM performance.