A Systematic Review and Quantitative Meta-Analysis of Multi-Modal SLAM Methods

Abstract

Simultaneous Localization and Mapping (SLAM) technology holds utmost significance in robotics, autonomous cars, and augmented reality. This paper first addresses the limitations of single-modal SLAM under dynamic environments (e.g., light changes, lack of texture, sensor degradation). The current paper aims to critically review and quantitatively evaluate multi-modal SLAM methods via a systematic review coupled with quantitative meta-analysis. This study details a comprehensive literature search strategy adopted herein, i.e., the databases queried (IEEE Xplore, ACM Digital Library, Scopus), keywords, and time frame. It further extracts the most critical information derived from the literature, such as sensor configurations, fusion architectures, and performance measures including absolute trajectory error (ATE), relative pose error (RPE), and real-time performance (FPS) on public datasets (KITTI, EuRoC). Finally, it outlines the key conclusions of the meta-analysis. For example, quantitative results illustrate that tightly coupled visual-inertial (V-I) methods exhibit high accuracy in high-speed motion scenes, while visual-LiDAR fusion systems demonstrate greater robustness for large-scale mapping tasks. The conclusion emphasizes that multi-modal fusion is an inexorable trend towards achieving high robustness and high accuracy for SLAM.