RynnUnify: Towards Unified Embodied Brain

One-line takeaway

RynnUnify achieves state-of-the-art embodied understanding and planning, with agent and robot control extensions demonstrating strong real-world performance.

RynnUnify 是一个16B统一具身模型，融合视觉语言推理实现长时域规划与物理预测，在多个任务中表现领先，并支持代理决策和机器人控制。

Key points

RynnUnify presents a 16B unified embodied model that combines video-language reasoning for long-horizon visual planning and physical prediction. It introduces an agent extension via reinforcement learning to reason about unobserved states, and a world-action module for robot control. Evaluated on Rynn-VP and Rynn-GEU benchmarks, it outperforms existing models in embodied understanding and planning, achieving high success rates in real-world robotic tasks.

• RynnUnify integrates video-language reasoning for long-horizon visual planning and physical prediction.
• Its agent extension improves unobserved state reasoning via reinforcement learning.
• World-action extension achieves 51.5% success on RoboCasa365 tasks and 46.7% on real-robot benchmarks.
• It establishes a unified foundation for embodied understanding, planning, and robot control.

Written only from the official arXiv abstract by an open-source model, then checked automatically (every number and name must appear in the abstract). The paper is authoritative.