SciGroveBeta
Robotics

RynnBrain 1.1: Towards More Capable and Generalizable Embodied Foundation Model

Kehan Li, Bohan Hou

Featured July 24, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

This paper teaches robots to grasp objects precisely and understand their 3D position by adding special training, and lets one smart robot brain control many different types of robots using a unified action space and smart masking.

In depth
This paper introduces RynnBrain 1.1, an embodied foundation model family that significantly enhances robot manipulation capabilities. It achieves this by incorporating novel pretraining tasks like contact point prediction and native 3D grounding, which provide more precise and physically aligned representations. Furthermore, the development of a unified cross-embodiment action space with embodiment-specific masking enables a single policy to control diverse robots and benefits from joint multi-task and multi-embodiment training.

Key Takeaways

  • 1
    RynnBrain 1.1 integrates contact point prediction and native 3D grounding into its pretraining, yielding representations that are more directly aligned with robot manipulation tasks.
  • 2
    The model introduces a unified cross-embodiment action space combined with embodiment-specific masking, allowing a single policy to generalize across heterogeneous robot platforms and control interfaces.
  • 3
    A systematic scaling analysis reveals that embodied pretraining is critical for reasoning-intensive tasks, transforming negative scaling observed in general VLMs into consistent performance gains with increased model capacity.

Conceptual Flow

HIGH LEVEL
1
Methodology: Unified Robot Brain

The robot brain takes in what it sees and hears, then figures out what to do by picking actions from a big list, only using the actions its body can actually make.

Robot Sees
Robot Hears
Robot Brain Thinks
Big List of Actions
2
Results: Smarter, More Flexible Robots

By learning how to grasp and understand 3D space, the new robot brain helps different robots work better together and perform complex tasks more reliably than older methods.

Old Robot Brain
New Robot Brain
Compare Performance
Better Grasping
Better 3D Sense
Works on Many Robots