XEmbodied: A Foundation Model with Enhanced Geometric and Physical Cues for Large-Scale Embodied Environments
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910149149458432 |
|---|---|
| author | Qian, Kangan Xie, ChuChu Zhong, Yang Pang, Jingrui Jiao, Siwen Jiang, Sicong Huang, Zilin Wang, Yunlong Jiang, Kun Yang, Mengmeng Ye, Hao Zhang, Guanghao Ye, Hangjun Chen, Guang Chen, Long Yang, Diange |
| author_facet | Qian, Kangan Xie, ChuChu Zhong, Yang Pang, Jingrui Jiao, Siwen Jiang, Sicong Huang, Zilin Wang, Yunlong Jiang, Kun Yang, Mengmeng Ye, Hao Zhang, Guanghao Ye, Hangjun Chen, Guang Chen, Long Yang, Diange |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_18484 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | XEmbodied: A Foundation Model with Enhanced Geometric and Physical Cues for Large-Scale Embodied Environments Qian, Kangan Xie, ChuChu Zhong, Yang Pang, Jingrui Jiao, Siwen Jiang, Sicong Huang, Zilin Wang, Yunlong Jiang, Kun Yang, Mengmeng Ye, Hao Zhang, Guanghao Ye, Hangjun Chen, Guang Chen, Long Yang, Diange Computer Vision and Pattern Recognition Multimedia Robotics 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. |
| title | XEmbodied: A Foundation Model with Enhanced Geometric and Physical Cues for Large-Scale Embodied Environments |
| topic | Computer Vision and Pattern Recognition Multimedia Robotics |
| url | https://arxiv.org/abs/2604.18484 |