InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909848491261952 |
|---|---|
| author | Chen, Xinyi Chen, Yilun Fu, Yanwei Gao, Ning Jia, Jiaya Jin, Weiyang Li, Hao Mu, Yao Pang, Jiangmiao Qiao, Yu Tian, Yang Wang, Bin Wang, Bolun Wang, Fangjing Wang, Hanqing Wang, Tai Wang, Ziqin Wei, Xueyuan Wu, Chao Yang, Shuai Ye, Jinhui Yu, Junqiu Zeng, Jia Zhang, Jingjing Zhang, Jinyu Zhang, Shi Zheng, Feng Zhou, Bowen Zhu, Yangkun |
| author_facet | Chen, Xinyi Chen, Yilun Fu, Yanwei Gao, Ning Jia, Jiaya Jin, Weiyang Li, Hao Mu, Yao Pang, Jiangmiao Qiao, Yu Tian, Yang Wang, Bin Wang, Bolun Wang, Fangjing Wang, Hanqing Wang, Tai Wang, Ziqin Wei, Xueyuan Wu, Chao Yang, Shuai Ye, Jinhui Yu, Junqiu Zeng, Jia Zhang, Jingjing Zhang, Jinyu Zhang, Shi Zheng, Feng Zhou, Bowen Zhu, Yangkun |
| contents | We introduce InternVLA-M1, a unified framework for spatial grounding and robot control that advances instruction-following robots toward scalable, general-purpose intelligence. Its core idea is spatially guided vision-language-action training, where spatial grounding serves as the critical link between instructions and robot actions. InternVLA-M1 employs a two-stage pipeline: (i) spatial grounding pre-training on over 2.3M spatial reasoning data to determine ``where to act'' by aligning instructions with visual, embodiment-agnostic positions, and (ii) spatially guided action post-training to decide ``how to act'' by generating embodiment-aware actions through plug-and-play spatial prompting. This spatially guided training recipe yields consistent gains: InternVLA-M1 outperforms its variant without spatial guidance by +14.6% on SimplerEnv Google Robot, +17% on WidowX, and +4.3% on LIBERO Franka, while demonstrating stronger spatial reasoning capability in box, point, and trace prediction. To further scale instruction following, we built a simulation engine to collect 244K generalizable pick-and-place episodes, enabling a 6.2% average improvement across 200 tasks and 3K+ objects. In real-world clustered pick-and-place, InternVLA-M1 improved by 7.3%, and with synthetic co-training, achieved +20.6% on unseen objects and novel configurations. Moreover, in long-horizon reasoning-intensive scenarios, it surpassed existing works by over 10%. These results highlight spatially guided training as a unifying principle for scalable and resilient generalist robots. Code and models are available at https://github.com/InternRobotics/InternVLA-M1. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_13778 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy Chen, Xinyi Chen, Yilun Fu, Yanwei Gao, Ning Jia, Jiaya Jin, Weiyang Li, Hao Mu, Yao Pang, Jiangmiao Qiao, Yu Tian, Yang Wang, Bin Wang, Bolun Wang, Fangjing Wang, Hanqing Wang, Tai Wang, Ziqin Wei, Xueyuan Wu, Chao Yang, Shuai Ye, Jinhui Yu, Junqiu Zeng, Jia Zhang, Jingjing Zhang, Jinyu Zhang, Shi Zheng, Feng Zhou, Bowen Zhu, Yangkun Robotics Artificial Intelligence Computer Vision and Pattern Recognition We introduce InternVLA-M1, a unified framework for spatial grounding and robot control that advances instruction-following robots toward scalable, general-purpose intelligence. Its core idea is spatially guided vision-language-action training, where spatial grounding serves as the critical link between instructions and robot actions. InternVLA-M1 employs a two-stage pipeline: (i) spatial grounding pre-training on over 2.3M spatial reasoning data to determine ``where to act'' by aligning instructions with visual, embodiment-agnostic positions, and (ii) spatially guided action post-training to decide ``how to act'' by generating embodiment-aware actions through plug-and-play spatial prompting. This spatially guided training recipe yields consistent gains: InternVLA-M1 outperforms its variant without spatial guidance by +14.6% on SimplerEnv Google Robot, +17% on WidowX, and +4.3% on LIBERO Franka, while demonstrating stronger spatial reasoning capability in box, point, and trace prediction. To further scale instruction following, we built a simulation engine to collect 244K generalizable pick-and-place episodes, enabling a 6.2% average improvement across 200 tasks and 3K+ objects. In real-world clustered pick-and-place, InternVLA-M1 improved by 7.3%, and with synthetic co-training, achieved +20.6% on unseen objects and novel configurations. Moreover, in long-horizon reasoning-intensive scenarios, it surpassed existing works by over 10%. These results highlight spatially guided training as a unifying principle for scalable and resilient generalist robots. Code and models are available at https://github.com/InternRobotics/InternVLA-M1. |
| title | InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy |
| topic | Robotics Artificial Intelligence Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2510.13778 |