The rapid evolution toward next-generation embodied AI has driven remarkable convergence across computer vision, graphics, and robotics, with growing emphasis on robust 3D understanding for in-the-wild and real-world applications. Recent breakthroughs have introduced powerful 3D foundational models that demonstrate unprecedented zero-shot generalization across diverse 3D reconstruction tasks. Meanwhile, advances in dynamic reconstruction representations and vision-language-action (VLA) models have enabled practical real-to-sim-to-real transfer, language-grounded robotic manipulation, and navigation in real-world settings.