The Year in AI Papers
The year/Independent research

Paper 2607.17977

RynnBrain 1.1: Towards More Capable and Generalizable Embodied Foundation Model

Kehan Li, Bohan Hou, Minghao Zhu, Tianyi Zhang, Zesen Cheng, Zhikai Wang, Sicong Leng, Xin Li, Xiao Lin, Biying Yao, Minghua Zeng, Jiangpin Liu, et al.

Published
Jul 2026
Research lab
Independent
Citations
2
GitHub
Not linked

01 In brief

Summary

RynnBrain 1.1 is a family of embodied foundation models (2B, 9B, 122B-A10B) built on Qwen3.5, introducing contact-point prediction and native 3D grounding (for 2B/9B) to better align with robot manipulation.

The 122B-A10B model outperforms all evaluated proprietary and open-source models on VSI-Bench, MMSI, and RefSpatial-Bench.

Scaling analysis shows embodied pretraining is crucial for reasoning-intensive tasks, where Qwen3.5 exhibits negative scaling.

RynnBrain-VLA, using a unified 81-dimensional action space with embodiment-specific masking, is deployed on Unitree G1, Astribot-S1, and Tianji-Wuji.

Real-robot experiments show RynnBrain-initialized policies outperform Qwen-based and generalist VLAs (GR00T N1.7, π0.5), with joint multi-task and multi-embodiment training improving average process score from 91.28% to 94.14% and success rate from 86.67% to 91.67% over per-task training.

The 9B model achieves 41.12 AP@15 on SUN RGB-D and 23.44 AP3D on WildDet3D-Bench, surpassing specialized detectors.

Contact-point prediction uses a compact (p, θ) representation, trained on 2.6M samples from multiple grasp datasets.

The training pipeline includes unified spatio-temporal representation, physically grounded output space, and a data mixture covering general MLLM, cognition, localization, 3D grounding, contact points, and planning data.

Real-time Chunking (RTC) enables smooth low-latency control with a 32-step action chunk and new inference every 5 steps.

02 From the paper

Abstract

We present RynnBrain 1.1, a family of embodied foundation models spanning 2B, 9B, and 122B-A10B scales. Trained with a unified spatio-temporal and physically grounded framework, RynnBrain 1.1 supports embodied perception, spatial reasoning, localization, and planning. Compared with RynnBrain 1.0, it further introduces contact-point prediction across the model family and native 3D grounding for the 2B and 9B models, yielding representations and outputs that are more directly aligned with robot manipulation. We also develop RynnBrain-VLA with a unified cross-embodiment action space and embodiment-specific masking, and deploy it on Unitree G1, Astribot-S1, and Tianji-Wuji. RynnBrain 1.1 achieves strong results on embodied cognition, localization, and 3D grounding, with the 122B-A10B model outperforming all evaluated proprietary and open-source models on VSI-Bench, MMSI, and RefSpatial-Bench. Real-robot experiments show that RynnBrain-initialized policies outperform Qwen-based and representative generalist VLAs, while joint multi-task and multi-embodiment training improves process scores and success rates over per-task training.