The field of 3D scene understanding and reconstruction is rapidly advancing, with a focus on developing scalable and efficient methods for embodied semantic scene completion, sparse-view guided scene updates, and dynamic mesh modeling. Recent works have introduced innovative approaches, such as temporal Gaussian splatting and cross-temporal 3D Gaussian splatting, to address the challenges of reconstructing and updating 3D scenes from continuous egocentric observations or sparse images. Another significant direction is the integration of 2D priors from foundation models into unified 4D Gaussian Splatting representations, enabling more accurate and consistent scene understanding. Noteworthy papers in this area include: TGSFormer, which achieves state-of-the-art results on both local and embodied SSC benchmarks with superior accuracy and scalability. Cross-Temporal 3D Gaussian Splatting, which efficiently reconstructs and updates 3D scenes across different time periods using sparse images and previously captured scene priors. TagSplat, which introduces a topology-aware dynamic reconstruction framework based on Gaussian Splatting, enabling topology-consistent mesh sequences with high accuracy. Motion4D, which integrates 2D priors from foundation models into a unified 4D Gaussian Splatting representation, significantly outperforming existing approaches in scene understanding tasks. SyncTrack4D, which achieves sub-frame synchronization accuracy and high-fidelity 4D reconstruction for unsynchronized video sets without assuming predefined scene objects or prior models.