g2o vs. Ceres: Optimizing Scan Matching in Cartographer SLAM

TL;DR

This paper compares g2o and Ceres solvers in Cartographer SLAM, showing Ceres generally outperforms g2o in speed and accuracy, while g2o has advantages in obstacle detection.

Contribution

It provides a comprehensive performance comparison of g2o and Ceres within Cartographer SLAM, highlighting their strengths and weaknesses.

Findings

Ceres converges faster and produces more accurate maps.

g2o excels in localized obstacle detection.

Ceres improves overall map clarity in real-world scenarios.

Abstract

This article presents a comparative analysis of g2o and Ceres solvers in enhancing scan matching performance within the Cartographer framework. Cartographer, a widely-used library for Simultaneous Localization and Mapping (SLAM), relies on optimization algorithms to refine pose estimates and improve map accuracy. The research aims to evaluate the performance, efficiency, and accuracy of the g2o solver in comparison to the Ceres solver, which is the default in Cartographer. In our experiments comparing Ceres and g2o within Cartographer, Ceres outperformed g2o in terms of speed, convergence efficiency, and overall map clarity. Ceres required fewer iterations and less time to converge, producing more accurate and well-defined maps, especially in real-world mapping scenarios with the AgileX LIMO robot. However, g2o excelled in localized obstacle detection, highlighting its value in specific situations.