In this work, we present a review of the recent method of 3D Gaussian Splatting (3D-GS), which explicitly represents the 3D scenes. This approach gives the impression of being a prospective alternative to the implicit Neural Radiance Field (NeRF) method for the task of novel-view synthesis with its higher rendering speed. We go through the fundamental principles behind this method, its representation, optimization process, and limitations. We also list several diverse applications of 3D-GS in robotics, including Simultaneous Localization and Mapping (SLAM), dynamic scene modeling, and 3D scene understanding.
Мохрат М. (науч. рук. Колюбин С.А.) A Review of 3D Gaussian Splatting Method and Its Applications In Robotics // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2024]. URL: https://kmu.itmo.ru/digests/article/12748