Resource Utilization of 2D SLAM Algorithms in ROS-Based Systems: an Empirical Evaluation

Simultaneous localization and mapping (SLAM) is an important task in robotic systems, which entails mapping an environment while keeping track of the robot’s position within the created map. The Robot Operating System (ROS) offers various packages for this functionality, where each one may lead to different performance and resource usage. Therefore, this study aims to investigate the impact of different ROS-based SLAM algorithms on resource utilization, including possible trade-offs with performance (e.g., the accuracy of the created map). The investigation is centered on primary experiments involving multiple runs of a single robot, which alternates between four SLAM algorithms: Cartographer, Gmapping, Hector SLAM, and Karto. During these experiments, the robot autonomously navigates through two types of arenas: point-to-point (multi-goal navigation) and circular (returning to the starting position after following the perimeter). Throughout these trials, the robot’s performance is assessed based on the ROS system’s efficiency and energy consumption. In a secondary set of experiments, the tests are repeated, but with key SLAM algorithm parameters reconfigured to evaluate their impact. The experiment results reveal Karto as the most efficient algorithm across all evaluated metrics, and the one that creates the most visually consistent maps. Cartographer was the algorithm that showed the least promising results regarding both, energy consumption and CPU utilization. Furthermore, Gmapping was the algorithm most susceptible to changes in SLAM algorithms’ parameter values. The results presented in this study are, therefore, key for resource-aware design choices when using SLAM algorithms in the context of ROS-based systems.