XEmbodied: A Foundation Model with Enhanced Geometric and Physical Cues for Large-Scale Embodied Environments

TL;DR

XEmbodied is a foundation model that enhances vision-language models with 3D geometric awareness and physical cues, improving reasoning and generalization in embodied environment tasks.

Contribution

It introduces a structured 3D adapter and physical signal distillation to integrate geometry and physics into vision-language models for embodied AI.

Findings

Significantly improves spatial reasoning and traffic semantics.

Demonstrates robust performance across 18 benchmarks.

Enhances out-of-distribution generalization in embodied tasks.

Abstract

Vision-Language-Action (VLA) models drive next-generation autonomous systems, but training them requires scalable, high-quality annotations from complex environments. Current cloud pipelines rely on generic vision-language models (VLMs) that lack geometric reasoning and domain semantics due to their 2D image-text pretraining. To address this mismatch, we propose XEmbodied, a cloud-side foundation model that endows VLMs with intrinsic 3D geometric awareness and interaction with physical cues (e.g., occupancy grids, 3D boxes). Instead of treating geometry as auxiliary input, XEmbodied integrates geometric representations via a structured 3D Adapter and distills physical signals into context tokens using an Efficient Image-Embodied Adapter. Through progressive domain curriculum and reinforcement learning post-training, XEmbodied preserves general capabilities while demonstrating robust performance across 18 public benchmarks. It significantly improves spatial reasoning, traffic semantics, embodied affordance, and out-of-distribution generalization for large-scale scenario mining and embodied VQA.