A quantitative study of tuning ROS gmapping parameters and their effect on performing indoor 2D SLAM

Abstract

Simultaneous localization and mapping (SLAM) complexity reduction is a fast progressing research area. Its attraction is owed to the potential commercial benefits of developing low cost yet highly effective SLAM based robotic applications. ROS gmapping package offers a lightweight incorporation of FastSLAM 2.0. The package has been used with different ROS supported robotic platforms and showed remarkable success. However, the effect of the package mapping parameters seem not to be fully exploited, especially with low cost robotic platform with no full ROS support such as Hercules platform. This paper presents a full implementation and performance quantitative evaluation on the gmapping package running on both standard PC and Raspberry Pi processors. We study the effects of tuning the number of particles, the displacement update and the resampling threshold by separately varying each of these parameters to several incremental values and running the algorithm on a recorded dataset. For each run, a grid map was constructed and the performance was evaluated based on mapping accuracy, CPU load and memory consumption. We are then able to propose a tuning guidelines to enlighten the gmapping execution while maintaining high performance.