Moduslam: A Modular Framework for Factor Graph-Based Localization and Mapping

Modern autonomous vehicles and robots rely on versatile sensors for diverse localization and mapping tasks. Factor graphs serve as a powerful approach for consistent sensor fusion, enabling maximum a posteriori state estimation. Recent advancements in LiDAR and Radar point cloud mapping, visualinertial odometry, bundle adjustment, and rendering require the integration of complex methods into a unified pipeline. Such integration necessitates a software architecture that is modular, scalable, and easily modifiable. This paper addresses key software design challenges in localization and mapping frameworks. It introduces a hybrid architecture combining hierarchical tree-based and flat service-based structures to streamline development, testing, and application. The introduced data structures facilitate efficient module interaction, enabling seamless integration of new functionalities. The resulting design ensures flexibility and adaptivity, supporting various algorithms across different tasks while catering to a broad research community. The framework is designed to accommodate emerging trends, such as semantic mapping and hierarchical environment representations. Our open-source library, implemented in Python, is publicly available at https://github.com/fatrybl/ModuSLAM.