RynnBrain 1.1: Towards a More Capable and Generalizable Embodied Foundation Model

RynnBrain 1.1 introduces embodied foundation models at 2B, 9B, and 122B-A10B scales for perception, spatial reasoning, localization, and planning. The update adds contact-point prediction across the family and native 3D grounding for the 2B and 9B models. Its RynnBrain-VLA uses a unified cross-embodiment action space with embodiment-specific masking, and has been deployed on Unitree G1, Astribot-S1, and Tianji-Wuji robots. The abstract reports that the 122B-A10B model outperforms evaluated proprietary and open-source models on VSI-Bench, MMSI, and RefSpatial-Bench, while initialized policies improve real-robot results over Qwen-based and other generalist VLAs.