The field of 3D scene understanding and autonomous driving is moving towards more efficient and robust learning paradigms. Researchers are exploring innovative approaches to address the challenges of high computational costs, scarce high-quality annotated datasets, and domain shift issues. One notable direction is the development of data-centric frameworks that emphasize enhancing data quality and training efficiency. These frameworks are designed to filter out noisy samples, reduce dependence on large-scale labeled data, and improve the overall performance of 3D scene understanding models. Another significant area of research is the development of novel multi-view projection frameworks for domain generalization and adaptation in 3D semantic segmentation. These frameworks are capable of aligning Lidar scans, rendering them from multiple virtual camera poses, and generating high-quality 2D datasets for training. Test-time adaptation is also an active area of research, with a focus on developing lightweight and scalable model merging frameworks that can efficiently adapt to dynamic and unpredictable test-time conditions. Noteworthy papers in this area include: DC-Scene, which proposes a data-centric framework for 3D scene understanding that achieves state-of-the-art performance while reducing training costs. seg_3D_by_PC2D, which introduces a novel multi-view projection framework for domain generalization and adaptation in 3D semantic segmentation, achieving state-of-the-art results in UDA and close to state-of-the-art in DG. CodeMerge, which presents a lightweight and scalable model merging framework for robust test-time adaptation in autonomous driving, achieving strong performance across challenging benchmarks. Extremely Simple Multimodal Outlier Synthesis, which proposes a simple and fast method for multimodal outlier synthesis, achieving state-of-the-art performance with a significant speedup.