Towards a Framework for SLAM Performance Investigation on Mobile Robots

Abstract

Simultaneous Localization And Mapping (SLAM) is the process of estimating the location of a mobile robot concurrently with constructing a model (the map) of the surrounding environment. The SLAM community has made breathtaking progress over the last two decades, varying from indoor mobile robot navigation to large-scale real-world applications. Many SLAM algorithms have been proposed in both research and applied problems. However, choosing the most suitable one among the SLAM algorithms and building it on a specific robot platform is not always easy. In this study, we develop a framework that allows us to deploy various SLAM algorithms for performance investigations. We perform experiments on two well-known SLAM algorithms, including Hector and Google Cartographer SLAMs, to show how to use the proposed framework for studying SLAM performance in terms of accuracy, processing time, and hardware resource consumption (i.e., CPU and RAM utilization). The experimental results show that Hector SLAM has better accuracy performance than Cartographer. Meanwhile, Cartographer has outstanding time performance but consuming more CPUs and RAM as a trade-off. These results are also valuable references for comparison with other studies.