ORBSLAM3

ORB-SLAM3 is a complete simultaneous localization and mapping (SLAM) system that supports monocular, stereo, and RGB-D cameras with inertial measurement unit (IMU) data. Developed by Raul Mur-Artal and collaborators, it represents the next generation of the ORB-SLAM family and is distributed as open-source software. The system is designed for real-time operation on standard CPUs and is widely used in robotics, augmented reality, and autonomous systems.

The architecture is built around a multi-threaded pipeline that includes tracking, local mapping, loop closing, and relocalization. ORB features are extracted and used to create a map of 3D points and keyframes, with place recognition performed through a Bag-of-Words approach to detect revisits. Global and local optimizations are performed through pose graph optimization and local bundle adjustment to produce accurate camera trajectories and map estimates.

A key feature of ORB-SLAM3 is its tightly coupled visual-inertial SLAM capability. IMU measurements are integrated with visual data through preintegration, enabling robust scale estimation (for monocular setups), improved robustness to fast motions, and more accurate pose, velocity, gravity direction, and IMU bias estimation. The system supports initialization, relocalization, loop closure, and map merging across multiple sessions, providing metric-scale trajectories and consistent maps across different sensors and runs.

ORB-SLAM3 is available as open-source software and has been adopted as a reference implementation for visual-inertial SLAM research and applications. It builds upon the concepts of its predecessors while expanding multi-sensor integration and robustness.