RynnBrain 1.1: Towards More Capable and Generalizable Embodied Foundation Model
The paper says RynnBrain 1.1 adds robot-oriented grounding and action modeling across multiple model sizes and embodiments.
The model family spans 2B, 9B, and 122B-A10B scales, with support for perception, spatial reasoning, localization, and planning. Compared with RynnBrain 1.0, it adds contact-point prediction across the family and native 3D grounding for the 2B and 9B versions. The authors also introduce RynnBrain-VLA, using a unified cross-embodiment action space, and report deployments on Unitree G1, Astribot-S1, and Tianji-Wuji. They claim the largest model leads evaluated proprietary and open-source systems on VSI-Bench, MMSI, and RefSpatial-Bench, while real-robot tests show stronger policy performance than Qwen-based and other generalist VLAs. HF Daily Papers' note
The model family spans 2B, 9B, and 122B-A10B scales, with support for perception, spatial reasoning, localization, and planning. Compared with RynnBrain 1.0, it adds contact-point prediction across the family and native 3D grounding for the 2B and 9B versions. The authors also introduce RynnBrain-VLA, using a unified cross-embodiment action space, and report deployments on Unitree G1, Astribot-S1, and Tianji-Wuji. They claim the largest model leads evaluated proprietary and open-source systems on VSI-Bench, MMSI, and RefSpatial-Bench, while real-robot tests show stronger policy performance than Qwen-based and other generalist VLAs. HF Daily Papers' note
score 6