The field of embodied intelligence has seen a major player emerge. Today, Alibaba DAMO Academy officially launched RynnBrain, an embodied intelligence brain foundation model, and open-sourced seven models in the full series, including a 30B MoE scale, to the developer community. This move marks a solid step forward for domestic embodied brain base models in the open-source ecosystem.

image.png

According to the introduction, the core breakthrough of RynnBrain lies in being the first to endow robots with spatiotemporal memory and spatial reasoning capabilities. Through the deep integration of these two key capabilities, the level of intelligent interaction of robots has made a qualitative leap, enabling them to better understand physical space and perform complex instructions. In the latest 16 open-source embodied intelligence evaluation lists, RynnBrain set new records (SOTA), with performance that has surpassed international top models such as Google Gemini Robotics ER1.5.

This open source includes models of different parameter scales, aiming to adapt to various industrial and service robot scenarios. DAMO Academy stated that by open-sourcing the RynnBrain series models, it hopes to reduce the research and development threshold of embodied intelligence and accelerate the integration of robots into real life.

Project: https://github.com/alibaba-damo-academy/RynnBrain