Selective scan matching techniques for enhanced graph slam in autonomous underwater vehicle localization

Abstract

This thesis investigates the performance of four prominent scan-matching algorithms— Iterative Closest Point (ICP), Generalized Iterative Closest Point (GICP), Gaussian Mix ture Model (GMM), and Probabilistic Iterative Correspondence (pIC)—within the frame work of Graph Simultaneous Localization and Mapping (Graph SLAM), particularly fo cusing on underwater environments. These environments pose unique challenges due to inherent noise in sensor data and dynamic underwater conditions. The study evalu ates how these algorithms influence the accuracy and reliability of SLAM in mapping and navigation. The findings demonstrate that while ICP offers improvements over basic dead reck oning, GICP and pIC significantly enhance the fidelity of SLAM maps and trajectory accuracy, attributed to their advanced handling of noise and alignment errors. A critical aspect of this research was examining the role of uncertainty estimation in scan match ing, where pIC’s capability to directly estimate uncertainty proved beneficial for effective loop closure and error minimization. However, the absence of inherent uncertainty esti mation in ICP and GICP necessitates external covariance estimation, which can lead to suboptimal corrections if inaccurately applied. The thesis underscores the necessity of integrating robust loop closure mechanisms 125 and accurate covariance models in SLAM systems, especially for long-term deployments in complex environments. Future work should explore algorithmic enhancements, hy brid approaches combining multiple scan-matching techniques, and the development of adaptive covariance models that respond to real-time environmental feedback. Through these advancements, this research aims to pave the way for more sophisticated and re liable autonomous navigation systems in underwater and other challenging operational contexts.

9