2026. Physically Grounded Multimodal World Models for Embodied Robot Interaction via 3D Representation Alignment. Journal of Advanced Artificial Intelligence Research. 1, 1 (Jul. 2026).