Improving Orb-Slam3 Performance in Textureless Environments: Mapping and Localization in Hospital Corridors

Abstract

Simultaneous Localization and Mapping (SLAM) plays a pivotal role in robotics applications, particularly for autonomous navigation in indoor environments. However, hospital corridors present unique challenges for traditional SLAM systems due to their textureless surfaces and homogeneous appearances. This study proposes an enhanced visual SLAM framework that integrates machine learning-based feature extraction (DISK) and matching (LightGlue) into ORB-SLAM3 to address these challenges. Extensive experiments conducted in Toyota Memorial Hospital demonstrate significant improvements in both mapping and localization performance. In mapping tasks, our system achieves 100 % success rate compared to ORB-SLAM3's 80 %, while reducing mapping error by 13.9 %. For localization tasks, our system maintains robust performance across different trajectories, achieving a 90.3 % tracking rate compared to ORBSLAM3's 23.5% in challenging scenarios. The localization error is reduced by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 1. 1 \%}$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 0. 9 \%}$</tex> in same-trajectory and differenttrajectory scenarios, respectively. These results demonstrate that our enhanced framework successfully addresses the challenges of visual SLAM in hospital environments, providing a more reliable solution for autonomous robot navigation in healthcare facilities.