RTABMap

RTAB-Map (Real-Time Appearance-Based Mapping) is an open-source SLAM library designed for real-time 2D and 3D mapping with visual and depth sensors. It is commonly used in robotics and augmented reality to build spatial maps and estimate trajectories while recognizing previously seen places. The system emphasizes long-term mapping by managing memory of past observations to support loop closures.

At its core, RTAB-Map maintains a memory management scheme that separates short-term and long-term observations to enable scalable mapping. Place recognition uses a bag-of-words approach to detect loops, and a pose graph optimizer (e.g., g2o) computes consistent camera trajectories after loop closures. The result is a coherent 3D map and a labeled pose graph.

RTAB-Map primarily consumes RGB-D and stereo camera data but can also process 3D point clouds from other sensors via ROS; it outputs maps as point clouds, meshes, and occupancy grids and can export trajectories. It supports multi-session mapping and streaming operation, with options for offline processing.

The library is written in C++, cross-platform, and available as a ROS package as well as a standalone library. It integrates with DBow (bag-of-words) for place recognition and PCL for point cloud processing; the project is released under a permissive license and maintained by a community of researchers.

RTAB-Map has been widely adopted for autonomous navigation, indoor mapping, and AR/VR research. Its memory-based approach to loop closure makes it suitable for long-term mapping in changing environments, and its ROS integration facilitates deployment on mobile robots and drones.