Core Insights - Alibaba's Damo Academy has released the RynnBrain foundational model for embodied intelligence, which includes seven models, such as the 30B MoE, and has been fully open-sourced [1] - The RynnBrain model introduces temporal memory and physical space reasoning capabilities to robots, allowing them to record and recall historical states during task execution [1] - In 16 open-source evaluations of embodied intelligence, RynnBrain outperformed existing models like Google's Gemini Robotics ER1.5 and NVIDIA's Cosmos Reason2 in various tasks [1] Group 1 - The RynnBrain model enables robots to remember the time and spatial state of a task when interrupted, allowing them to resume previous tasks seamlessly [1] - Damo Academy's embodied intelligence lab head, Zhao Deli, stated that RynnBrain represents a significant step towards the development of general embodied intelligence [1] - Damo Academy has previously open-sourced related models such as WorldVLA and RynnEC, and has introduced the RynnRCP protocol to advance AI applications in real-world physical scenarios [1]
阿里达摩院开源具身大脑基模RynnBrain,首次让机器人拥有时空记忆