# ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM

**Authors:** Carlos Campos, Richard Elvira, Juan J. Gómez Rodríguez, José M. M. Montiel, Juan D. Tardós  
**arXiv:** [2007.11898](https://arxiv.org/abs/2007.11898)  
**Date:** 2020-07-23  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2007.11898v2/

## Key points

- ORB-SLAM3 is an advanced open-source library for simultaneous localization and mapping (SLAM) systems.
- It supports visual, visual-inertial, and multi-map SLAM using various camera types.
- The system operates in real time across different environments.
- ORB-SLAM3 achieves 2 to 5 times more accuracy compared to previous methods.
- It incorporates a multiple map system with improved recall capabilities.
- ORB-SLAM3 seamlessly merges new maps with previously mapped areas.
- Co-visible keyframes are included in bundle adjustment computations for enhanced accuracy.
- Experimental evaluations show that ORB-SLAM3 performs robustly and surpasses the accuracy levels of state-of-the-art systems.
- The stereo inertial SLAM implementation of ORB-SLAM3 is suitable for applications in augmented reality and virtual reality scenarios.
- The source code of ORB-SLAM3 is publicly available for researchers and practitioners.

## Summary

ORB-SLAM3 is an advanced open-source library that introduces several novel features to enhance the accuracy and robustness of simultaneous localization and mapping (SLAM) systems. It supports visual, visual-inertial, and multi-map SLAM using various camera types such as monocular, stereo, and RGB-D cameras with both pinhole and fisheye lens models. The first major innovation of ORB-SLAM3 is its tightly integrated visual-inertial SLAM system that relies on Maximum a Posteriori (MAP) estimation throughout the entire process even during the initialization phase of the inertial measurement unit (IMU). This approach enables the system to operate in real time across different environments from small indoor spaces to large outdoor areas. Furthermore, ORB-SLAM3 achieves a significant improvement in accuracy compared to previous methods with results showing it to be 2 to 5 times more accurate. Another key feature introduced by ORB-SLAM3 is its multiple map system which incorporates a new place recognition method with improved recall capabilities. As a result, ORB-SLAM3 can effectively handle long periods of poor visual information. When the system loses track or encounters challenging conditions it initiates a new map that seamlessly merges with previously mapped areas upon revisiting them. This capability allows ORB-SLAM3 to reuse all previous information at every stage of the algorithm instead of relying solely on recent data like traditional visual odometry systems. By including co visible keyframes in bundle adjustment computations which provide high parallax observations even if they are temporally distant or from previous mapping sessions ORB-SLAM3 achieves enhanced accuracy. Experimental evaluations demonstrate that ORB-SLAM3 performs as robustly as state of the art systems available in literature across all sensor configurations while significantly surpassing their accuracy levels. Notably, the stereo inertial SLAM implementation of ORB-SLAM3 achieves an average accuracy of 3.6 cm on the EuRoC drone dataset and 9 mm under quick hand held motions in the TUM VI room dataset making it suitable for applications in augmented reality (AR) and virtual reality (VR) scenarios. To benefit the research community, the authors have made the source code of ORB SLAM3 publicly available providing a valuable resource for researchers and practitioners working on visual, visual inertial and multi map SLAM systems.