Visual-Based Simultaneous Localization and Mapping (VSLAM) Techniques for Robots: A Scientific Review

Abstract

The main problem facing autonomous robots is to navigate in an environment with the ability to determine its location and simultaneously build a map, SLAM technique can formulate this requirement efficiently. In this paper Filter-based, Graph-based, and AI-based Visual-SLAM techniques have been reviewed. The review shows that the first method suffers from high computations when the number of landmarks increases. The Graph-based algorithms are exposed to drift-error problems which cause a delocalization and require optimization. The AI-based vSLAM has the advantage of not-having complicated mathematical models in the algorithm, and it shows an efficient performance in various environments. The reviewed algorithms utilize different cameras including mono, stereo, and RGB-D cameras. The low-cost RGB-D cameras encourage implementation in modern autonomous robots. This work introduces a scientific-based overview of vSLAM to the reader, by explaining all phases of SLAM, the state-of-the-art algorithms, highlighting the strengths and weaknesses of each paradigm.