RynnBrain

Abstract

We present RynnBrain, an embodied foundation model grounded in physical reality. Moving beyond passive observation, RynnBrain anchors its understanding in the physical world through comprehensive egocentric cognition, precise spatiotemporal grounding, and real task planning. This systematic upgrade enables active, physics-aware reasoning and complex task execution.

We release RynnBrain in dense (2B, 8B) and MoE (30B) variants, alongside three specialized models: RynnBrain‑Plan (manipulation planning), RynnBrain‑Nav (navigation), and RynnBrain‑CoP (spatial reasoning).

🌟 Key Highlights

Comprehensive Egocentric Understanding:
Excels in fine-grained video understanding and egocentric cognition, covering diverse tasks such as embodied QA, counting, and OCR.
Diverse Spatiotemporal Localization:
Possesses powerful localization capabilities across episodic memory, enabling precise identification of objects, target areas, and motion trajectories.
Physical-Space Reasoning:
Interleaves textual and spatial grounding to anchor reasoning firmly in physical reality.
Physics-Aware Planning:
Integrates localized affordances and object info into planning, enabling downstream VLA models to execute intricate tasks.

Overview

Results

Click on the image to view a larger version.

2B & 8B Results

30B-A3B Results

Planning Results

Navigation Results

Skill Demonstration

Showcasing specific capabilities.

Put it back in its original position

Demonstrated capabilities: Object memory and spatial memory

3 Gods Assembly (三仙归洞)

Demonstrated capabilities: Object memory ability and complex video understanding capabilities

Sorting and Disruption

Demonstrated capabilities: Object classification and planning capabilities, and the capability to perform multiple tasks simultaneously.

Looking for low-calorie foods

Common sense understanding and OCR capabilities

Fun portioned bread

Spatial planning and long-horizon planning capabilities

Real-Robot Demos

Click on the video to view a larger version.

Task 1

Planning on getting food

Task 2

Planning on dividing food

Task-ood-1

Sorting food into categories

Task-ood-2

Sorting food in lines

Navigation Demos

Click on the video to view a larger version.

Navigation Task 1

Go forward toward the windows. Go toward the the far couch, Stop next to the couch, in front of the windows.

Navigation Task 2

Walk towards fireplace, walk on the left side of the brown couch, turn right after couch, turn left after four person table and stop in front of barstools.

Navigation Task 3

Go straight pas the couches and the fireplace on the right. Go through the double doors. Turn right and pass the couches. Go into the room on the right. Wait near the round table and chairs with the white flower in the table.

Navigation Task 4

Walk into the bedroom toward the fireplace. Go around the fireplace on the left. Stop in the doorway before you reach the rug.

Navigation Task 5

Walk past the couch and left through the room into the hallway. Walk past the microwave out of the room and stop.

Navigation Task 6

Walk through the door and over the overpass. Wait on the white circular rug.

Model Architecture

RynnBrain Bench

BibTeX

@article{rynnbrain2026,
  title   = {RynnBrain: Open Embodied Foundation Models},
  author  = {Ronghao Dang and Jiayan Guo and Bohan Hou and Sicong Leng and Kehan Li and Xin Li and Jiangping Liu and Yunxuan Mao and Zhikai Wang and Yuqian Yuan and Minghao Zhu and Xiao Lin and Yang Bai and Qian Jiang and Yaxi Zhao and Minghua Zeng and Junlong Gao and Yuming Jiang and Jun Cen and Siteng Huang and Liuyi Wang and Wenqiao Zhang and Chengju Liu and Jianfei Yang and Shijian Lu and Deli Zhao},
  year    = {2026}
}