OmniVLA-RL: A Vision-Language-Action Model with Spatial Understanding and Online RL
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866910161632755712 |
|---|---|
| author | Jie, Haoxiang Yan, Yaoyuan Wei, Xiangyu Wang, Kailin Yan, Hongjie Heng, Zhiyou Chen, Daocheng |
| author_facet | Jie, Haoxiang Yan, Yaoyuan Wei, Xiangyu Wang, Kailin Yan, Hongjie Heng, Zhiyou Chen, Daocheng |
| contents | Visual-Language-Action (VLA) models represent a paradigm shift in embodied AI, yet existing frameworks often struggle with imprecise spatial perception, suboptimal multimodal fusion, and instability in reinforcement learning. To bridge these gaps, we propose OmniVLA-RL, a novel architecture that leverages a Mix-of-Transformers (MoT) design to synergistically integrate reasoning, spatial, and action experts. Furthermore, we introduce Flow-GSPO, which reformulates flow matching as a Stochastic Differential Equation (SDE) process and integrates it with Group Segmented Policy Optimization (GSPO) to enhance action precision and training robustness. Extensive evaluations on the LIBERO and LIBERO-Plus benchmarks demonstrate that OmniVLA-RL achieves decent overall performance and surpasses mainstream existing methods, effectively overcoming the fundamental limitations of current VLA models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_17706 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | OmniVLA-RL: A Vision-Language-Action Model with Spatial Understanding and Online RL Jie, Haoxiang Yan, Yaoyuan Wei, Xiangyu Wang, Kailin Yan, Hongjie Heng, Zhiyou Chen, Daocheng Robotics Visual-Language-Action (VLA) models represent a paradigm shift in embodied AI, yet existing frameworks often struggle with imprecise spatial perception, suboptimal multimodal fusion, and instability in reinforcement learning. To bridge these gaps, we propose OmniVLA-RL, a novel architecture that leverages a Mix-of-Transformers (MoT) design to synergistically integrate reasoning, spatial, and action experts. Furthermore, we introduce Flow-GSPO, which reformulates flow matching as a Stochastic Differential Equation (SDE) process and integrates it with Group Segmented Policy Optimization (GSPO) to enhance action precision and training robustness. Extensive evaluations on the LIBERO and LIBERO-Plus benchmarks demonstrate that OmniVLA-RL achieves decent overall performance and surpasses mainstream existing methods, effectively overcoming the fundamental limitations of current VLA models. |
| title | OmniVLA-RL: A Vision-Language-Action Model with Spatial Understanding and Online RL |
| topic | Robotics |
| url | https://arxiv.org/abs/2604.17706 |