Year: 2024
Type: conference-paper
Abstract: In this study, we propose a novel methodology for Graduated Non-Convexity (GNC) and validate its effectiveness through its application in robust pose graph optimization, which is a critical element in SLAM (Simultaneous Localization and Mapping) backends. Conventional GNC techniques typically depend on heuristic strategies for determining the GNC schedule, specifically in the adjustment of the control parameter µ to increase non-convexity. In contrast, our approach utilizes the characteristics of convex functions and convex optimization to ascertain the boundary points where convexity can no longer be assured. This advancement allows for the elimination of superfluous optimization iterations present in traditional methods, thereby improving both the efficiency and robustness of the process. We provide evidence that our method surpasses the current leading techniques in terms of both speed and accuracy when applied to robust back-end pose graph optimization utilizing GNC. Furthermore, our work builds upon and enhances the open-source riSAM framework. The code is available at: https://github.com/SNU-DLLAB/EGNC-PGO.
Authors: Wonseok Kang, Jaehyun Kim, Jiseong Chung, Seungwon Choi, Tae-Wan Kim
Institution: Seoul National University
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
Funder: Ministry of Oceans and Fisheries
Award: G8501051832
OpenAlex ID: W4405180314