Multi-Objective Optimisation of RTAB-Map Parameters Using Genetic Algorithm for Indoor 2D SLAM

Abstract

Currently, there are multiple packages available to implement different Simultaneous Localisation And Mapping (SLAM) approaches in Robot Operating System (ROS). To effectively obtain sensor data, these packages use parameters whose values are set from prior knowledge and experience working with robots and SLAM. In this research, using a Multi-Objective Genetic Algorithm (MOGA) to optimise the values for these parameters is proposed. Using MOGA allows trade-offs between the objectives using Pareto dominance technique. Three parameters from the RTAB-Map package are considered for optimisation using three different MOGA mechanisms, Dominance Count, Dominance Rank and Switching Fitness. The quality of the map generated for every set of parameters is taken as the indicator of its performance. The number of corners, number of contours and the proportion of occupied cells in the map are used as quantitative measures of map quality. Finally, results obtained from the algorithm are tested on a Quanser QBot2 robot.