**Dynin-Robotics: Omnimodal Unified Diffusion Vision-Language-Action Model** >

**Dynin-Robotics: Omnimodal Unified Diffusion Vision-Language-Action Model**

> Visual goal and dynamics prediction can provide language-conditioned robot policies with both a target outcome and a representation of action-dependent scene changes. We bring these predictions into action generation and selection through a shared trajectory model. Dynin-Robotics implements this formulation on Dynin-Omni, an omnimodal masked-diffusion backbone, representing language, visual observations, goals, and a...

**来源信息**
- **来源**：HuggingFace Daily Papers（社区热门论文）
- **分类**：论文
- **发布时间**：2026-09-15 08:00（北京时间）
- **原文**：[打开原文](https://huggingface.co/papers/2609.13053)