XEmbodied: A Foundation Model with Enhanced Geometric and Physical Cues for Large-Scale Embodied Environments
XEmbodied is a foundation model that enhances Vision-Language-Action models by integrating intrinsic 3D geometric awareness and physical cues from environments. It achieves this through a 3D Adapter for geometry and an Image-Embodied Adapter for physical data, trained via a progressive domain curriculum. The result is a model capable of more robust spatial reasoning and interaction in embodied tasks.