The field of robotic manipulation is witnessing a significant shift towards the integration of vision-language models to enhance spatial awareness and adaptability. Researchers are exploring innovative approaches to bridge the gap between high-level task semantics and low-level geometric features, enabling robots to better understand their environment and perform complex tasks. Notably, the development of frameworks that incorporate 3D representations, spatial-temporal understanding, and reconstructive vision-language-action models is advancing the field. These advancements have the potential to improve robotic manipulation in various scenarios, including dynamic warehousing and real-world applications. Noteworthy papers include: PASG, which introduces a closed-loop framework for automated geometric primitive extraction and semantic anchoring. GeoVLA, which effectively integrates 3D information to advance robotic manipulation. ReconVLA, which proposes a reconstructive VLA model with an implicit grounding paradigm to guide visual attention grounding on the correct target.